]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super-intel.c
projects / thirdparty / mdadm.git / blob
? search:


541365965448819c8db4dd696212f64a858e7fab
[thirdparty/mdadm.git] / super-intel.c
1 /*
2 * mdadm - Intel(R) Matrix Storage Manager Support
3 *
4 * Copyright (C) 2002-2008 Intel Corporation
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define HAVE_STDINT_H 1
21 #include "mdadm.h"
22 #include "mdmon.h"
23 #include "sha1.h"
24 #include "platform-intel.h"
25 #include <values.h>
26 #include <scsi/sg.h>
27 #include <ctype.h>
28 #include <dirent.h>
29
30 /* MPB == Metadata Parameter Block */
31 #define MPB_SIGNATURE "Intel Raid ISM Cfg Sig. "
32 #define MPB_SIG_LEN (strlen(MPB_SIGNATURE))
33 #define MPB_VERSION_RAID0 "1.0.00"
34 #define MPB_VERSION_RAID1 "1.1.00"
35 #define MPB_VERSION_MANY_VOLUMES_PER_ARRAY "1.2.00"
36 #define MPB_VERSION_3OR4_DISK_ARRAY "1.2.01"
37 #define MPB_VERSION_RAID5 "1.2.02"
38 #define MPB_VERSION_5OR6_DISK_ARRAY "1.2.04"
39 #define MPB_VERSION_CNG "1.2.06"
40 #define MPB_VERSION_ATTRIBS "1.3.00"
41 #define MAX_SIGNATURE_LENGTH 32
42 #define MAX_RAID_SERIAL_LEN 16
43
44 #define MPB_ATTRIB_CHECKSUM_VERIFY __cpu_to_le32(0x80000000)
45 #define MPB_ATTRIB_PM __cpu_to_le32(0x40000000)
46 #define MPB_ATTRIB_2TB __cpu_to_le32(0x20000000)
47 #define MPB_ATTRIB_RAID0 __cpu_to_le32(0x00000001)
48 #define MPB_ATTRIB_RAID1 __cpu_to_le32(0x00000002)
49 #define MPB_ATTRIB_RAID10 __cpu_to_le32(0x00000004)
50 #define MPB_ATTRIB_RAID1E __cpu_to_le32(0x00000008)
51 #define MPB_ATTRIB_RAID5 __cpu_to_le32(0x00000010)
52 #define MPB_ATTRIB_RAIDCNG __cpu_to_le32(0x00000020)
53
54 #define MPB_SECTOR_CNT 418
55 #define IMSM_RESERVED_SECTORS 4096
56
57 /* Disk configuration info. */
58 #define IMSM_MAX_DEVICES 255
59 struct imsm_disk {
60 __u8 serial[MAX_RAID_SERIAL_LEN];/* 0xD8 - 0xE7 ascii serial number */
61 __u32 total_blocks; /* 0xE8 - 0xEB total blocks */
62 __u32 scsi_id; /* 0xEC - 0xEF scsi ID */
63 #define SPARE_DISK __cpu_to_le32(0x01) /* Spare */
64 #define CONFIGURED_DISK __cpu_to_le32(0x02) /* Member of some RaidDev */
65 #define FAILED_DISK __cpu_to_le32(0x04) /* Permanent failure */
66 #define USABLE_DISK __cpu_to_le32(0x08) /* Fully usable unless FAILED_DISK is set */
67 __u32 status; /* 0xF0 - 0xF3 */
68 __u32 owner_cfg_num; /* which config 0,1,2... owns this disk */
69 #define IMSM_DISK_FILLERS 4
70 __u32 filler[IMSM_DISK_FILLERS]; /* 0xF4 - 0x107 MPB_DISK_FILLERS for future expansion */
71 };
72
73 /* RAID map configuration infos. */
74 struct imsm_map {
75 __u32 pba_of_lba0; /* start address of partition */
76 __u32 blocks_per_member;/* blocks per member */
77 __u32 num_data_stripes; /* number of data stripes */
78 __u16 blocks_per_strip;
79 __u8 map_state; /* Normal, Uninitialized, Degraded, Failed */
80 #define IMSM_T_STATE_NORMAL 0
81 #define IMSM_T_STATE_UNINITIALIZED 1
82 #define IMSM_T_STATE_DEGRADED 2
83 #define IMSM_T_STATE_FAILED 3
84 __u8 raid_level;
85 #define IMSM_T_RAID0 0
86 #define IMSM_T_RAID1 1
87 #define IMSM_T_RAID5 5 /* since metadata version 1.2.02 ? */
88 __u8 num_members; /* number of member disks */
89 __u8 num_domains; /* number of parity domains */
90 __u8 failed_disk_num; /* valid only when state is degraded */
91 __u8 reserved[1];
92 __u32 filler[7]; /* expansion area */
93 #define IMSM_ORD_REBUILD (1 << 24)
94 __u32 disk_ord_tbl[1]; /* disk_ord_tbl[num_members],
95 * top byte contains some flags
96 */
97 } __attribute__ ((packed));
98
99 struct imsm_vol {
100 __u32 curr_migr_unit;
101 __u32 checkpoint_id; /* id to access curr_migr_unit */
102 __u8 migr_state; /* Normal or Migrating */
103 #define MIGR_INIT 0
104 #define MIGR_REBUILD 1
105 #define MIGR_VERIFY 2 /* analagous to echo check > sync_action */
106 #define MIGR_GEN_MIGR 3
107 #define MIGR_STATE_CHANGE 4
108 __u8 migr_type; /* Initializing, Rebuilding, ... */
109 __u8 dirty;
110 __u8 fs_state; /* fast-sync state for CnG (0xff == disabled) */
111 __u16 verify_errors; /* number of mismatches */
112 __u16 bad_blocks; /* number of bad blocks during verify */
113 __u32 filler[4];
114 struct imsm_map map[1];
115 /* here comes another one if migr_state */
116 } __attribute__ ((packed));
117
118 struct imsm_dev {
119 __u8 volume[MAX_RAID_SERIAL_LEN];
120 __u32 size_low;
121 __u32 size_high;
122 #define DEV_BOOTABLE __cpu_to_le32(0x01)
123 #define DEV_BOOT_DEVICE __cpu_to_le32(0x02)
124 #define DEV_READ_COALESCING __cpu_to_le32(0x04)
125 #define DEV_WRITE_COALESCING __cpu_to_le32(0x08)
126 #define DEV_LAST_SHUTDOWN_DIRTY __cpu_to_le32(0x10)
127 #define DEV_HIDDEN_AT_BOOT __cpu_to_le32(0x20)
128 #define DEV_CURRENTLY_HIDDEN __cpu_to_le32(0x40)
129 #define DEV_VERIFY_AND_FIX __cpu_to_le32(0x80)
130 #define DEV_MAP_STATE_UNINIT __cpu_to_le32(0x100)
131 #define DEV_NO_AUTO_RECOVERY __cpu_to_le32(0x200)
132 #define DEV_CLONE_N_GO __cpu_to_le32(0x400)
133 #define DEV_CLONE_MAN_SYNC __cpu_to_le32(0x800)
134 #define DEV_CNG_MASTER_DISK_NUM __cpu_to_le32(0x1000)
135 __u32 status; /* Persistent RaidDev status */
136 __u32 reserved_blocks; /* Reserved blocks at beginning of volume */
137 __u8 migr_priority;
138 __u8 num_sub_vols;
139 __u8 tid;
140 __u8 cng_master_disk;
141 __u16 cache_policy;
142 __u8 cng_state;
143 __u8 cng_sub_state;
144 #define IMSM_DEV_FILLERS 10
145 __u32 filler[IMSM_DEV_FILLERS];
146 struct imsm_vol vol;
147 } __attribute__ ((packed));
148
149 struct imsm_super {
150 __u8 sig[MAX_SIGNATURE_LENGTH]; /* 0x00 - 0x1F */
151 __u32 check_sum; /* 0x20 - 0x23 MPB Checksum */
152 __u32 mpb_size; /* 0x24 - 0x27 Size of MPB */
153 __u32 family_num; /* 0x28 - 0x2B Checksum from first time this config was written */
154 __u32 generation_num; /* 0x2C - 0x2F Incremented each time this array's MPB is written */
155 __u32 error_log_size; /* 0x30 - 0x33 in bytes */
156 __u32 attributes; /* 0x34 - 0x37 */
157 __u8 num_disks; /* 0x38 Number of configured disks */
158 __u8 num_raid_devs; /* 0x39 Number of configured volumes */
159 __u8 error_log_pos; /* 0x3A */
160 __u8 fill[1]; /* 0x3B */
161 __u32 cache_size; /* 0x3c - 0x40 in mb */
162 __u32 orig_family_num; /* 0x40 - 0x43 original family num */
163 __u32 pwr_cycle_count; /* 0x44 - 0x47 simulated power cycle count for array */
164 __u32 bbm_log_size; /* 0x48 - 0x4B - size of bad Block Mgmt Log in bytes */
165 #define IMSM_FILLERS 35
166 __u32 filler[IMSM_FILLERS]; /* 0x4C - 0xD7 RAID_MPB_FILLERS */
167 struct imsm_disk disk[1]; /* 0xD8 diskTbl[numDisks] */
168 /* here comes imsm_dev[num_raid_devs] */
169 /* here comes BBM logs */
170 } __attribute__ ((packed));
171
172 #define BBM_LOG_MAX_ENTRIES 254
173
174 struct bbm_log_entry {
175 __u64 defective_block_start;
176 #define UNREADABLE 0xFFFFFFFF
177 __u32 spare_block_offset;
178 __u16 remapped_marked_count;
179 __u16 disk_ordinal;
180 } __attribute__ ((__packed__));
181
182 struct bbm_log {
183 __u32 signature; /* 0xABADB10C */
184 __u32 entry_count;
185 __u32 reserved_spare_block_count; /* 0 */
186 __u32 reserved; /* 0xFFFF */
187 __u64 first_spare_lba;
188 struct bbm_log_entry mapped_block_entries[BBM_LOG_MAX_ENTRIES];
189 } __attribute__ ((__packed__));
190
191
192 #ifndef MDASSEMBLE
193 static char *map_state_str[] = { "normal", "uninitialized", "degraded", "failed" };
194 #endif
195
196 static unsigned int sector_count(__u32 bytes)
197 {
198 return ((bytes + (512-1)) & (~(512-1))) / 512;
199 }
200
201 static unsigned int mpb_sectors(struct imsm_super *mpb)
202 {
203 return sector_count(__le32_to_cpu(mpb->mpb_size));
204 }
205
206 struct intel_dev {
207 struct imsm_dev *dev;
208 struct intel_dev *next;
209 int index;
210 };
211
212 /* internal representation of IMSM metadata */
213 struct intel_super {
214 union {
215 void *buf; /* O_DIRECT buffer for reading/writing metadata */
216 struct imsm_super *anchor; /* immovable parameters */
217 };
218 size_t len; /* size of the 'buf' allocation */
219 void *next_buf; /* for realloc'ing buf from the manager */
220 size_t next_len;
221 int updates_pending; /* count of pending updates for mdmon */
222 int creating_imsm; /* flag to indicate container creation */
223 int current_vol; /* index of raid device undergoing creation */
224 __u32 create_offset; /* common start for 'current_vol' */
225 struct intel_dev *devlist;
226 struct dl {
227 struct dl *next;
228 int index;
229 __u8 serial[MAX_RAID_SERIAL_LEN];
230 int major, minor;
231 char *devname;
232 struct imsm_disk disk;
233 int fd;
234 int extent_cnt;
235 struct extent *e; /* for determining freespace @ create */
236 int raiddisk; /* slot to fill in autolayout */
237 } *disks;
238 struct dl *add; /* list of disks to add while mdmon active */
239 struct dl *missing; /* disks removed while we weren't looking */
240 struct bbm_log *bbm_log;
241 const char *hba; /* device path of the raid controller for this metadata */
242 const struct imsm_orom *orom; /* platform firmware support */
243 };
244
245 struct extent {
246 unsigned long long start, size;
247 };
248
249 /* definition of messages passed to imsm_process_update */
250 enum imsm_update_type {
251 update_activate_spare,
252 update_create_array,
253 update_add_disk,
254 };
255
256 struct imsm_update_activate_spare {
257 enum imsm_update_type type;
258 struct dl *dl;
259 int slot;
260 int array;
261 struct imsm_update_activate_spare *next;
262 };
263
264 struct disk_info {
265 __u8 serial[MAX_RAID_SERIAL_LEN];
266 };
267
268 struct imsm_update_create_array {
269 enum imsm_update_type type;
270 int dev_idx;
271 struct imsm_dev dev;
272 };
273
274 struct imsm_update_add_disk {
275 enum imsm_update_type type;
276 };
277
278 static struct supertype *match_metadata_desc_imsm(char *arg)
279 {
280 struct supertype *st;
281
282 if (strcmp(arg, "imsm") != 0 &&
283 strcmp(arg, "default") != 0
284 )
285 return NULL;
286
287 st = malloc(sizeof(*st));
288 memset(st, 0, sizeof(*st));
289 st->ss = &super_imsm;
290 st->max_devs = IMSM_MAX_DEVICES;
291 st->minor_version = 0;
292 st->sb = NULL;
293 return st;
294 }
295
296 #ifndef MDASSEMBLE
297 static __u8 *get_imsm_version(struct imsm_super *mpb)
298 {
299 return &mpb->sig[MPB_SIG_LEN];
300 }
301 #endif
302
303 /* retrieve a disk directly from the anchor when the anchor is known to be
304 * up-to-date, currently only at load time
305 */
306 static struct imsm_disk *__get_imsm_disk(struct imsm_super *mpb, __u8 index)
307 {
308 if (index >= mpb->num_disks)
309 return NULL;
310 return &mpb->disk[index];
311 }
312
313 #ifndef MDASSEMBLE
314 /* retrieve a disk from the parsed metadata */
315 static struct imsm_disk *get_imsm_disk(struct intel_super *super, __u8 index)
316 {
317 struct dl *d;
318
319 for (d = super->disks; d; d = d->next)
320 if (d->index == index)
321 return &d->disk;
322
323 return NULL;
324 }
325 #endif
326
327 /* generate a checksum directly from the anchor when the anchor is known to be
328 * up-to-date, currently only at load or write_super after coalescing
329 */
330 static __u32 __gen_imsm_checksum(struct imsm_super *mpb)
331 {
332 __u32 end = mpb->mpb_size / sizeof(end);
333 __u32 *p = (__u32 *) mpb;
334 __u32 sum = 0;
335
336 while (end--) {
337 sum += __le32_to_cpu(*p);
338 p++;
339 }
340
341 return sum - __le32_to_cpu(mpb->check_sum);
342 }
343
344 static size_t sizeof_imsm_map(struct imsm_map *map)
345 {
346 return sizeof(struct imsm_map) + sizeof(__u32) * (map->num_members - 1);
347 }
348
349 struct imsm_map *get_imsm_map(struct imsm_dev *dev, int second_map)
350 {
351 struct imsm_map *map = &dev->vol.map[0];
352
353 if (second_map && !dev->vol.migr_state)
354 return NULL;
355 else if (second_map) {
356 void *ptr = map;
357
358 return ptr + sizeof_imsm_map(map);
359 } else
360 return map;
361
362 }
363
364 /* return the size of the device.
365 * migr_state increases the returned size if map[0] were to be duplicated
366 */
367 static size_t sizeof_imsm_dev(struct imsm_dev *dev, int migr_state)
368 {
369 size_t size = sizeof(*dev) - sizeof(struct imsm_map) +
370 sizeof_imsm_map(get_imsm_map(dev, 0));
371
372 /* migrating means an additional map */
373 if (dev->vol.migr_state)
374 size += sizeof_imsm_map(get_imsm_map(dev, 1));
375 else if (migr_state)
376 size += sizeof_imsm_map(get_imsm_map(dev, 0));
377
378 return size;
379 }
380
381 #ifndef MDASSEMBLE
382 /* retrieve disk serial number list from a metadata update */
383 static struct disk_info *get_disk_info(struct imsm_update_create_array *update)
384 {
385 void *u = update;
386 struct disk_info *inf;
387
388 inf = u + sizeof(*update) - sizeof(struct imsm_dev) +
389 sizeof_imsm_dev(&update->dev, 0);
390
391 return inf;
392 }
393 #endif
394
395 static struct imsm_dev *__get_imsm_dev(struct imsm_super *mpb, __u8 index)
396 {
397 int offset;
398 int i;
399 void *_mpb = mpb;
400
401 if (index >= mpb->num_raid_devs)
402 return NULL;
403
404 /* devices start after all disks */
405 offset = ((void *) &mpb->disk[mpb->num_disks]) - _mpb;
406
407 for (i = 0; i <= index; i++)
408 if (i == index)
409 return _mpb + offset;
410 else
411 offset += sizeof_imsm_dev(_mpb + offset, 0);
412
413 return NULL;
414 }
415
416 static struct imsm_dev *get_imsm_dev(struct intel_super *super, __u8 index)
417 {
418 struct intel_dev *dv;
419
420 if (index >= super->anchor->num_raid_devs)
421 return NULL;
422 for (dv = super->devlist; dv; dv = dv->next)
423 if (dv->index == index)
424 return dv->dev;
425 return NULL;
426 }
427
428 static __u32 get_imsm_ord_tbl_ent(struct imsm_dev *dev, int slot)
429 {
430 struct imsm_map *map;
431
432 if (dev->vol.migr_state)
433 map = get_imsm_map(dev, 1);
434 else
435 map = get_imsm_map(dev, 0);
436
437 /* top byte identifies disk under rebuild */
438 return __le32_to_cpu(map->disk_ord_tbl[slot]);
439 }
440
441 #define ord_to_idx(ord) (((ord) << 8) >> 8)
442 static __u32 get_imsm_disk_idx(struct imsm_dev *dev, int slot)
443 {
444 __u32 ord = get_imsm_ord_tbl_ent(dev, slot);
445
446 return ord_to_idx(ord);
447 }
448
449 static void set_imsm_ord_tbl_ent(struct imsm_map *map, int slot, __u32 ord)
450 {
451 map->disk_ord_tbl[slot] = __cpu_to_le32(ord);
452 }
453
454 static int get_imsm_disk_slot(struct imsm_map *map, int idx)
455 {
456 int slot;
457 __u32 ord;
458
459 for (slot = 0; slot < map->num_members; slot++) {
460 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
461 if (ord_to_idx(ord) == idx)
462 return slot;
463 }
464
465 return -1;
466 }
467
468 static int get_imsm_raid_level(struct imsm_map *map)
469 {
470 if (map->raid_level == 1) {
471 if (map->num_members == 2)
472 return 1;
473 else
474 return 10;
475 }
476
477 return map->raid_level;
478 }
479
480 static int cmp_extent(const void *av, const void *bv)
481 {
482 const struct extent *a = av;
483 const struct extent *b = bv;
484 if (a->start < b->start)
485 return -1;
486 if (a->start > b->start)
487 return 1;
488 return 0;
489 }
490
491 static int count_memberships(struct dl *dl, struct intel_super *super)
492 {
493 int memberships = 0;
494 int i;
495
496 for (i = 0; i < super->anchor->num_raid_devs; i++) {
497 struct imsm_dev *dev = get_imsm_dev(super, i);
498 struct imsm_map *map = get_imsm_map(dev, 0);
499
500 if (get_imsm_disk_slot(map, dl->index) >= 0)
501 memberships++;
502 }
503
504 return memberships;
505 }
506
507 static struct extent *get_extents(struct intel_super *super, struct dl *dl)
508 {
509 /* find a list of used extents on the given physical device */
510 struct extent *rv, *e;
511 int i;
512 int memberships = count_memberships(dl, super);
513 __u32 reservation = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
514
515 rv = malloc(sizeof(struct extent) * (memberships + 1));
516 if (!rv)
517 return NULL;
518 e = rv;
519
520 for (i = 0; i < super->anchor->num_raid_devs; i++) {
521 struct imsm_dev *dev = get_imsm_dev(super, i);
522 struct imsm_map *map = get_imsm_map(dev, 0);
523
524 if (get_imsm_disk_slot(map, dl->index) >= 0) {
525 e->start = __le32_to_cpu(map->pba_of_lba0);
526 e->size = __le32_to_cpu(map->blocks_per_member);
527 e++;
528 }
529 }
530 qsort(rv, memberships, sizeof(*rv), cmp_extent);
531
532 /* determine the start of the metadata
533 * when no raid devices are defined use the default
534 * ...otherwise allow the metadata to truncate the value
535 * as is the case with older versions of imsm
536 */
537 if (memberships) {
538 struct extent *last = &rv[memberships - 1];
539 __u32 remainder;
540
541 remainder = __le32_to_cpu(dl->disk.total_blocks) -
542 (last->start + last->size);
543 /* round down to 1k block to satisfy precision of the kernel
544 * 'size' interface
545 */
546 remainder &= ~1UL;
547 /* make sure remainder is still sane */
548 if (remainder < ROUND_UP(super->len, 512) >> 9)
549 remainder = ROUND_UP(super->len, 512) >> 9;
550 if (reservation > remainder)
551 reservation = remainder;
552 }
553 e->start = __le32_to_cpu(dl->disk.total_blocks) - reservation;
554 e->size = 0;
555 return rv;
556 }
557
558 /* try to determine how much space is reserved for metadata from
559 * the last get_extents() entry, otherwise fallback to the
560 * default
561 */
562 static __u32 imsm_reserved_sectors(struct intel_super *super, struct dl *dl)
563 {
564 struct extent *e;
565 int i;
566 __u32 rv;
567
568 /* for spares just return a minimal reservation which will grow
569 * once the spare is picked up by an array
570 */
571 if (dl->index == -1)
572 return MPB_SECTOR_CNT;
573
574 e = get_extents(super, dl);
575 if (!e)
576 return MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
577
578 /* scroll to last entry */
579 for (i = 0; e[i].size; i++)
580 continue;
581
582 rv = __le32_to_cpu(dl->disk.total_blocks) - e[i].start;
583
584 free(e);
585
586 return rv;
587 }
588
589 #ifndef MDASSEMBLE
590 static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
591 {
592 __u64 sz;
593 int slot;
594 struct imsm_map *map = get_imsm_map(dev, 0);
595 __u32 ord;
596
597 printf("\n");
598 printf("[%.16s]:\n", dev->volume);
599 printf(" UUID : %s\n", uuid);
600 printf(" RAID Level : %d\n", get_imsm_raid_level(map));
601 printf(" Members : %d\n", map->num_members);
602 slot = get_imsm_disk_slot(map, disk_idx);
603 if (slot >= 0) {
604 ord = get_imsm_ord_tbl_ent(dev, slot);
605 printf(" This Slot : %d%s\n", slot,
606 ord & IMSM_ORD_REBUILD ? " (out-of-sync)" : "");
607 } else
608 printf(" This Slot : ?\n");
609 sz = __le32_to_cpu(dev->size_high);
610 sz <<= 32;
611 sz += __le32_to_cpu(dev->size_low);
612 printf(" Array Size : %llu%s\n", (unsigned long long)sz,
613 human_size(sz * 512));
614 sz = __le32_to_cpu(map->blocks_per_member);
615 printf(" Per Dev Size : %llu%s\n", (unsigned long long)sz,
616 human_size(sz * 512));
617 printf(" Sector Offset : %u\n",
618 __le32_to_cpu(map->pba_of_lba0));
619 printf(" Num Stripes : %u\n",
620 __le32_to_cpu(map->num_data_stripes));
621 printf(" Chunk Size : %u KiB\n",
622 __le16_to_cpu(map->blocks_per_strip) / 2);
623 printf(" Reserved : %d\n", __le32_to_cpu(dev->reserved_blocks));
624 printf(" Migrate State : %s", dev->vol.migr_state ? "migrating" : "idle");
625 if (dev->vol.migr_state)
626 printf(": %s", dev->vol.migr_type ? "rebuilding" : "initializing");
627 printf("\n");
628 printf(" Map State : %s", map_state_str[map->map_state]);
629 if (dev->vol.migr_state) {
630 struct imsm_map *map = get_imsm_map(dev, 1);
631 printf(" <-- %s", map_state_str[map->map_state]);
632 }
633 printf("\n");
634 printf(" Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
635 }
636
637 static void print_imsm_disk(struct imsm_super *mpb, int index, __u32 reserved)
638 {
639 struct imsm_disk *disk = __get_imsm_disk(mpb, index);
640 char str[MAX_RAID_SERIAL_LEN + 1];
641 __u32 s;
642 __u64 sz;
643
644 if (index < 0)
645 return;
646
647 printf("\n");
648 snprintf(str, MAX_RAID_SERIAL_LEN + 1, "%s", disk->serial);
649 printf(" Disk%02d Serial : %s\n", index, str);
650 s = disk->status;
651 printf(" State :%s%s%s%s\n", s&SPARE_DISK ? " spare" : "",
652 s&CONFIGURED_DISK ? " active" : "",
653 s&FAILED_DISK ? " failed" : "",
654 s&USABLE_DISK ? " usable" : "");
655 printf(" Id : %08x\n", __le32_to_cpu(disk->scsi_id));
656 sz = __le32_to_cpu(disk->total_blocks) - reserved;
657 printf(" Usable Size : %llu%s\n", (unsigned long long)sz,
658 human_size(sz * 512));
659 }
660
661 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info);
662
663 static void examine_super_imsm(struct supertype *st, char *homehost)
664 {
665 struct intel_super *super = st->sb;
666 struct imsm_super *mpb = super->anchor;
667 char str[MAX_SIGNATURE_LENGTH];
668 int i;
669 struct mdinfo info;
670 char nbuf[64];
671 __u32 sum;
672 __u32 reserved = imsm_reserved_sectors(super, super->disks);
673
674
675 snprintf(str, MPB_SIG_LEN, "%s", mpb->sig);
676 printf(" Magic : %s\n", str);
677 snprintf(str, strlen(MPB_VERSION_RAID0), "%s", get_imsm_version(mpb));
678 printf(" Version : %s\n", get_imsm_version(mpb));
679 printf(" Family : %08x\n", __le32_to_cpu(mpb->family_num));
680 printf(" Generation : %08x\n", __le32_to_cpu(mpb->generation_num));
681 getinfo_super_imsm(st, &info);
682 fname_from_uuid(st, &info, nbuf,'-');
683 printf(" UUID : %s\n", nbuf + 5);
684 sum = __le32_to_cpu(mpb->check_sum);
685 printf(" Checksum : %08x %s\n", sum,
686 __gen_imsm_checksum(mpb) == sum ? "correct" : "incorrect");
687 printf(" MPB Sectors : %d\n", mpb_sectors(mpb));
688 printf(" Disks : %d\n", mpb->num_disks);
689 printf(" RAID Devices : %d\n", mpb->num_raid_devs);
690 print_imsm_disk(mpb, super->disks->index, reserved);
691 if (super->bbm_log) {
692 struct bbm_log *log = super->bbm_log;
693
694 printf("\n");
695 printf("Bad Block Management Log:\n");
696 printf(" Log Size : %d\n", __le32_to_cpu(mpb->bbm_log_size));
697 printf(" Signature : %x\n", __le32_to_cpu(log->signature));
698 printf(" Entry Count : %d\n", __le32_to_cpu(log->entry_count));
699 printf(" Spare Blocks : %d\n", __le32_to_cpu(log->reserved_spare_block_count));
700 printf(" First Spare : %llx\n", __le64_to_cpu(log->first_spare_lba));
701 }
702 for (i = 0; i < mpb->num_raid_devs; i++) {
703 struct mdinfo info;
704 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
705
706 super->current_vol = i;
707 getinfo_super_imsm(st, &info);
708 fname_from_uuid(st, &info, nbuf, '-');
709 print_imsm_dev(dev, nbuf + 5, super->disks->index);
710 }
711 for (i = 0; i < mpb->num_disks; i++) {
712 if (i == super->disks->index)
713 continue;
714 print_imsm_disk(mpb, i, reserved);
715 }
716 }
717
718 static void brief_examine_super_imsm(struct supertype *st)
719 {
720 /* We just write a generic IMSM ARRAY entry */
721 struct mdinfo info;
722 char nbuf[64];
723 char nbuf1[64];
724 struct intel_super *super = st->sb;
725 int i;
726
727 if (!super->anchor->num_raid_devs)
728 return;
729
730 getinfo_super_imsm(st, &info);
731 fname_from_uuid(st, &info, nbuf,'-');
732 printf("ARRAY metadata=imsm auto=md UUID=%s\n", nbuf + 5);
733 for (i = 0; i < super->anchor->num_raid_devs; i++) {
734 struct imsm_dev *dev = get_imsm_dev(super, i);
735
736 super->current_vol = i;
737 getinfo_super_imsm(st, &info);
738 fname_from_uuid(st, &info, nbuf1,'-');
739 printf("ARRAY /dev/md/%.16s container=%s\n"
740 " member=%d auto=mdp UUID=%s\n",
741 dev->volume, nbuf + 5, i, nbuf1 + 5);
742 }
743 }
744
745 static void detail_super_imsm(struct supertype *st, char *homehost)
746 {
747 struct mdinfo info;
748 char nbuf[64];
749
750 getinfo_super_imsm(st, &info);
751 fname_from_uuid(st, &info, nbuf,'-');
752 printf("\n UUID : %s\n", nbuf + 5);
753 }
754
755 static void brief_detail_super_imsm(struct supertype *st)
756 {
757 struct mdinfo info;
758 char nbuf[64];
759 getinfo_super_imsm(st, &info);
760 fname_from_uuid(st, &info, nbuf,'-');
761 printf(" UUID=%s", nbuf + 5);
762 }
763
764 static int imsm_read_serial(int fd, char *devname, __u8 *serial);
765 static void fd2devname(int fd, char *name);
766
767 static int imsm_enumerate_ports(const char *hba_path, int port_count, int host_base, int verbose)
768 {
769 /* dump an unsorted list of devices attached to ahci, as well as
770 * non-connected ports
771 */
772 int hba_len = strlen(hba_path) + 1;
773 struct dirent *ent;
774 DIR *dir;
775 char *path = NULL;
776 int err = 0;
777 unsigned long port_mask = (1 << port_count) - 1;
778
779 if (port_count > sizeof(port_mask) * 8) {
780 if (verbose)
781 fprintf(stderr, Name ": port_count %d out of range\n", port_count);
782 return 2;
783 }
784
785 /* scroll through /sys/dev/block looking for devices attached to
786 * this hba
787 */
788 dir = opendir("/sys/dev/block");
789 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
790 int fd;
791 char model[64];
792 char vendor[64];
793 char buf[1024];
794 int major, minor;
795 char *device;
796 char *c;
797 int port;
798 int type;
799
800 if (sscanf(ent->d_name, "%d:%d", &major, &minor) != 2)
801 continue;
802 path = devt_to_devpath(makedev(major, minor));
803 if (!path)
804 continue;
805 if (!path_attached_to_hba(path, hba_path)) {
806 free(path);
807 path = NULL;
808 continue;
809 }
810
811 /* retrieve the scsi device type */
812 if (asprintf(&device, "/sys/dev/block/%d:%d/device/xxxxxxx", major, minor) < 0) {
813 if (verbose)
814 fprintf(stderr, Name ": failed to allocate 'device'\n");
815 err = 2;
816 break;
817 }
818 sprintf(device, "/sys/dev/block/%d:%d/device/type", major, minor);
819 if (load_sys(device, buf) != 0) {
820 if (verbose)
821 fprintf(stderr, Name ": failed to read device type for %s\n",
822 path);
823 err = 2;
824 free(device);
825 break;
826 }
827 type = strtoul(buf, NULL, 10);
828
829 /* if it's not a disk print the vendor and model */
830 if (!(type == 0 || type == 7 || type == 14)) {
831 vendor[0] = '\0';
832 model[0] = '\0';
833 sprintf(device, "/sys/dev/block/%d:%d/device/vendor", major, minor);
834 if (load_sys(device, buf) == 0) {
835 strncpy(vendor, buf, sizeof(vendor));
836 vendor[sizeof(vendor) - 1] = '\0';
837 c = (char *) &vendor[sizeof(vendor) - 1];
838 while (isspace(*c) || *c == '\0')
839 *c-- = '\0';
840
841 }
842 sprintf(device, "/sys/dev/block/%d:%d/device/model", major, minor);
843 if (load_sys(device, buf) == 0) {
844 strncpy(model, buf, sizeof(model));
845 model[sizeof(model) - 1] = '\0';
846 c = (char *) &model[sizeof(model) - 1];
847 while (isspace(*c) || *c == '\0')
848 *c-- = '\0';
849 }
850
851 if (vendor[0] && model[0])
852 sprintf(buf, "%.64s %.64s", vendor, model);
853 else
854 switch (type) { /* numbers from hald/linux/device.c */
855 case 1: sprintf(buf, "tape"); break;
856 case 2: sprintf(buf, "printer"); break;
857 case 3: sprintf(buf, "processor"); break;
858 case 4:
859 case 5: sprintf(buf, "cdrom"); break;
860 case 6: sprintf(buf, "scanner"); break;
861 case 8: sprintf(buf, "media_changer"); break;
862 case 9: sprintf(buf, "comm"); break;
863 case 12: sprintf(buf, "raid"); break;
864 default: sprintf(buf, "unknown");
865 }
866 } else
867 buf[0] = '\0';
868 free(device);
869
870 /* chop device path to 'host%d' and calculate the port number */
871 c = strchr(&path[hba_len], '/');
872 *c = '\0';
873 if (sscanf(&path[hba_len], "host%d", &port) == 1)
874 port -= host_base;
875 else {
876 if (verbose) {
877 *c = '/'; /* repair the full string */
878 fprintf(stderr, Name ": failed to determine port number for %s\n",
879 path);
880 }
881 err = 2;
882 break;
883 }
884
885 /* mark this port as used */
886 port_mask &= ~(1 << port);
887
888 /* print out the device information */
889 if (buf[0]) {
890 printf(" Port%d : - non-disk device (%s) -\n", port, buf);
891 continue;
892 }
893
894 fd = dev_open(ent->d_name, O_RDONLY);
895 if (fd < 0)
896 printf(" Port%d : - disk info unavailable -\n", port);
897 else {
898 fd2devname(fd, buf);
899 printf(" Port%d : %s", port, buf);
900 if (imsm_read_serial(fd, NULL, (__u8 *) buf) == 0)
901 printf(" (%s)\n", buf);
902 else
903 printf("()\n");
904 }
905 close(fd);
906 free(path);
907 path = NULL;
908 }
909 if (path)
910 free(path);
911 if (dir)
912 closedir(dir);
913 if (err == 0) {
914 int i;
915
916 for (i = 0; i < port_count; i++)
917 if (port_mask & (1 << i))
918 printf(" Port%d : - no device attached -\n", i);
919 }
920
921 return err;
922 }
923
924 static int detail_platform_imsm(int verbose, int enumerate_only)
925 {
926 /* There are two components to imsm platform support, the ahci SATA
927 * controller and the option-rom. To find the SATA controller we
928 * simply look in /sys/bus/pci/drivers/ahci to see if an ahci
929 * controller with the Intel vendor id is present. This approach
930 * allows mdadm to leverage the kernel's ahci detection logic, with the
931 * caveat that if ahci.ko is not loaded mdadm will not be able to
932 * detect platform raid capabilities. The option-rom resides in a
933 * platform "Adapter ROM". We scan for its signature to retrieve the
934 * platform capabilities. If raid support is disabled in the BIOS the
935 * option-rom capability structure will not be available.
936 */
937 const struct imsm_orom *orom;
938 struct sys_dev *list, *hba;
939 DIR *dir;
940 struct dirent *ent;
941 const char *hba_path;
942 int host_base = 0;
943 int port_count = 0;
944
945 if (enumerate_only) {
946 if (check_env("IMSM_NO_PLATFORM") || find_imsm_orom())
947 return 0;
948 return 2;
949 }
950
951 list = find_driver_devices("pci", "ahci");
952 for (hba = list; hba; hba = hba->next)
953 if (devpath_to_vendor(hba->path) == 0x8086)
954 break;
955
956 if (!hba) {
957 if (verbose)
958 fprintf(stderr, Name ": unable to find active ahci controller\n");
959 free_sys_dev(&list);
960 return 2;
961 } else if (verbose)
962 fprintf(stderr, Name ": found Intel SATA AHCI Controller\n");
963 hba_path = hba->path;
964 hba->path = NULL;
965 free_sys_dev(&list);
966
967 orom = find_imsm_orom();
968 if (!orom) {
969 if (verbose)
970 fprintf(stderr, Name ": imsm option-rom not found\n");
971 return 2;
972 }
973
974 printf(" Platform : Intel(R) Matrix Storage Manager\n");
975 printf(" Version : %d.%d.%d.%d\n", orom->major_ver, orom->minor_ver,
976 orom->hotfix_ver, orom->build);
977 printf(" RAID Levels :%s%s%s%s%s\n",
978 imsm_orom_has_raid0(orom) ? " raid0" : "",
979 imsm_orom_has_raid1(orom) ? " raid1" : "",
980 imsm_orom_has_raid1e(orom) ? " raid1e" : "",
981 imsm_orom_has_raid10(orom) ? " raid10" : "",
982 imsm_orom_has_raid5(orom) ? " raid5" : "");
983 printf(" Chunk Sizes :%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
984 imsm_orom_has_chunk(orom, 2) ? " 2k" : "",
985 imsm_orom_has_chunk(orom, 4) ? " 4k" : "",
986 imsm_orom_has_chunk(orom, 8) ? " 8k" : "",
987 imsm_orom_has_chunk(orom, 16) ? " 16k" : "",
988 imsm_orom_has_chunk(orom, 32) ? " 32k" : "",
989 imsm_orom_has_chunk(orom, 64) ? " 64k" : "",
990 imsm_orom_has_chunk(orom, 128) ? " 128k" : "",
991 imsm_orom_has_chunk(orom, 256) ? " 256k" : "",
992 imsm_orom_has_chunk(orom, 512) ? " 512k" : "",
993 imsm_orom_has_chunk(orom, 1024*1) ? " 1M" : "",
994 imsm_orom_has_chunk(orom, 1024*2) ? " 2M" : "",
995 imsm_orom_has_chunk(orom, 1024*4) ? " 4M" : "",
996 imsm_orom_has_chunk(orom, 1024*8) ? " 8M" : "",
997 imsm_orom_has_chunk(orom, 1024*16) ? " 16M" : "",
998 imsm_orom_has_chunk(orom, 1024*32) ? " 32M" : "",
999 imsm_orom_has_chunk(orom, 1024*64) ? " 64M" : "");
1000 printf(" Max Disks : %d\n", orom->tds);
1001 printf(" Max Volumes : %d\n", orom->vpa);
1002 printf(" I/O Controller : %s\n", hba_path);
1003
1004 /* find the smallest scsi host number to determine a port number base */
1005 dir = opendir(hba_path);
1006 for (ent = dir ? readdir(dir) : NULL; ent; ent = readdir(dir)) {
1007 int host;
1008
1009 if (sscanf(ent->d_name, "host%d", &host) != 1)
1010 continue;
1011 if (port_count == 0)
1012 host_base = host;
1013 else if (host < host_base)
1014 host_base = host;
1015
1016 if (host + 1 > port_count + host_base)
1017 port_count = host + 1 - host_base;
1018
1019 }
1020 if (dir)
1021 closedir(dir);
1022
1023 if (!port_count || imsm_enumerate_ports(hba_path, port_count,
1024 host_base, verbose) != 0) {
1025 if (verbose)
1026 fprintf(stderr, Name ": failed to enumerate ports\n");
1027 return 2;
1028 }
1029
1030 return 0;
1031 }
1032 #endif
1033
1034 static int match_home_imsm(struct supertype *st, char *homehost)
1035 {
1036 /* the imsm metadata format does not specify any host
1037 * identification information. We return -1 since we can never
1038 * confirm nor deny whether a given array is "meant" for this
1039 * host. We rely on compare_super and the 'family_num' field to
1040 * exclude member disks that do not belong, and we rely on
1041 * mdadm.conf to specify the arrays that should be assembled.
1042 * Auto-assembly may still pick up "foreign" arrays.
1043 */
1044
1045 return -1;
1046 }
1047
1048 static void uuid_from_super_imsm(struct supertype *st, int uuid[4])
1049 {
1050 /* The uuid returned here is used for:
1051 * uuid to put into bitmap file (Create, Grow)
1052 * uuid for backup header when saving critical section (Grow)
1053 * comparing uuids when re-adding a device into an array
1054 * In these cases the uuid required is that of the data-array,
1055 * not the device-set.
1056 * uuid to recognise same set when adding a missing device back
1057 * to an array. This is a uuid for the device-set.
1058 *
1059 * For each of these we can make do with a truncated
1060 * or hashed uuid rather than the original, as long as
1061 * everyone agrees.
1062 * In each case the uuid required is that of the data-array,
1063 * not the device-set.
1064 */
1065 /* imsm does not track uuid's so we synthesis one using sha1 on
1066 * - The signature (Which is constant for all imsm array, but no matter)
1067 * - the family_num of the container
1068 * - the index number of the volume
1069 * - the 'serial' number of the volume.
1070 * Hopefully these are all constant.
1071 */
1072 struct intel_super *super = st->sb;
1073
1074 char buf[20];
1075 struct sha1_ctx ctx;
1076 struct imsm_dev *dev = NULL;
1077
1078 sha1_init_ctx(&ctx);
1079 sha1_process_bytes(super->anchor->sig, MPB_SIG_LEN, &ctx);
1080 sha1_process_bytes(&super->anchor->family_num, sizeof(__u32), &ctx);
1081 if (super->current_vol >= 0)
1082 dev = get_imsm_dev(super, super->current_vol);
1083 if (dev) {
1084 __u32 vol = super->current_vol;
1085 sha1_process_bytes(&vol, sizeof(vol), &ctx);
1086 sha1_process_bytes(dev->volume, MAX_RAID_SERIAL_LEN, &ctx);
1087 }
1088 sha1_finish_ctx(&ctx, buf);
1089 memcpy(uuid, buf, 4*4);
1090 }
1091
1092 #if 0
1093 static void
1094 get_imsm_numerical_version(struct imsm_super *mpb, int *m, int *p)
1095 {
1096 __u8 *v = get_imsm_version(mpb);
1097 __u8 *end = mpb->sig + MAX_SIGNATURE_LENGTH;
1098 char major[] = { 0, 0, 0 };
1099 char minor[] = { 0 ,0, 0 };
1100 char patch[] = { 0, 0, 0 };
1101 char *ver_parse[] = { major, minor, patch };
1102 int i, j;
1103
1104 i = j = 0;
1105 while (*v != '\0' && v < end) {
1106 if (*v != '.' && j < 2)
1107 ver_parse[i][j++] = *v;
1108 else {
1109 i++;
1110 j = 0;
1111 }
1112 v++;
1113 }
1114
1115 *m = strtol(minor, NULL, 0);
1116 *p = strtol(patch, NULL, 0);
1117 }
1118 #endif
1119
1120 static int imsm_level_to_layout(int level)
1121 {
1122 switch (level) {
1123 case 0:
1124 case 1:
1125 return 0;
1126 case 5:
1127 case 6:
1128 return ALGORITHM_LEFT_ASYMMETRIC;
1129 case 10:
1130 return 0x102;
1131 }
1132 return UnSet;
1133 }
1134
1135 static void getinfo_super_imsm_volume(struct supertype *st, struct mdinfo *info)
1136 {
1137 struct intel_super *super = st->sb;
1138 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
1139 struct imsm_map *map = get_imsm_map(dev, 0);
1140 struct dl *dl;
1141
1142 for (dl = super->disks; dl; dl = dl->next)
1143 if (dl->raiddisk == info->disk.raid_disk)
1144 break;
1145 info->container_member = super->current_vol;
1146 info->array.raid_disks = map->num_members;
1147 info->array.level = get_imsm_raid_level(map);
1148 info->array.layout = imsm_level_to_layout(info->array.level);
1149 info->array.md_minor = -1;
1150 info->array.ctime = 0;
1151 info->array.utime = 0;
1152 info->array.chunk_size = __le16_to_cpu(map->blocks_per_strip) << 9;
1153 info->array.state = !dev->vol.dirty;
1154
1155 info->disk.major = 0;
1156 info->disk.minor = 0;
1157 if (dl) {
1158 info->disk.major = dl->major;
1159 info->disk.minor = dl->minor;
1160 }
1161
1162 info->data_offset = __le32_to_cpu(map->pba_of_lba0);
1163 info->component_size = __le32_to_cpu(map->blocks_per_member);
1164 memset(info->uuid, 0, sizeof(info->uuid));
1165
1166 if (map->map_state == IMSM_T_STATE_UNINITIALIZED || dev->vol.dirty)
1167 info->resync_start = 0;
1168 else if (dev->vol.migr_state)
1169 info->resync_start = __le32_to_cpu(dev->vol.curr_migr_unit);
1170 else
1171 info->resync_start = ~0ULL;
1172
1173 strncpy(info->name, (char *) dev->volume, MAX_RAID_SERIAL_LEN);
1174 info->name[MAX_RAID_SERIAL_LEN] = 0;
1175
1176 info->array.major_version = -1;
1177 info->array.minor_version = -2;
1178 sprintf(info->text_version, "/%s/%d",
1179 devnum2devname(st->container_dev),
1180 info->container_member);
1181 info->safe_mode_delay = 4000; /* 4 secs like the Matrix driver */
1182 uuid_from_super_imsm(st, info->uuid);
1183 }
1184
1185 /* check the config file to see if we can return a real uuid for this spare */
1186 static void fixup_container_spare_uuid(struct mdinfo *inf)
1187 {
1188 struct mddev_ident_s *array_list;
1189
1190 if (inf->array.level != LEVEL_CONTAINER ||
1191 memcmp(inf->uuid, uuid_match_any, sizeof(int[4])) != 0)
1192 return;
1193
1194 array_list = conf_get_ident(NULL);
1195
1196 for (; array_list; array_list = array_list->next) {
1197 if (array_list->uuid_set) {
1198 struct supertype *_sst; /* spare supertype */
1199 struct supertype *_cst; /* container supertype */
1200
1201 _cst = array_list->st;
1202 _sst = _cst->ss->match_metadata_desc(inf->text_version);
1203 if (_sst) {
1204 memcpy(inf->uuid, array_list->uuid, sizeof(int[4]));
1205 free(_sst);
1206 break;
1207 }
1208 }
1209 }
1210 }
1211
1212 static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
1213 {
1214 struct intel_super *super = st->sb;
1215 struct imsm_disk *disk;
1216 __u32 s;
1217
1218 if (super->current_vol >= 0) {
1219 getinfo_super_imsm_volume(st, info);
1220 return;
1221 }
1222
1223 /* Set raid_disks to zero so that Assemble will always pull in valid
1224 * spares
1225 */
1226 info->array.raid_disks = 0;
1227 info->array.level = LEVEL_CONTAINER;
1228 info->array.layout = 0;
1229 info->array.md_minor = -1;
1230 info->array.ctime = 0; /* N/A for imsm */
1231 info->array.utime = 0;
1232 info->array.chunk_size = 0;
1233
1234 info->disk.major = 0;
1235 info->disk.minor = 0;
1236 info->disk.raid_disk = -1;
1237 info->reshape_active = 0;
1238 info->array.major_version = -1;
1239 info->array.minor_version = -2;
1240 strcpy(info->text_version, "imsm");
1241 info->safe_mode_delay = 0;
1242 info->disk.number = -1;
1243 info->disk.state = 0;
1244 info->name[0] = 0;
1245
1246 if (super->disks) {
1247 __u32 reserved = imsm_reserved_sectors(super, super->disks);
1248
1249 disk = &super->disks->disk;
1250 info->data_offset = __le32_to_cpu(disk->total_blocks) - reserved;
1251 info->component_size = reserved;
1252 s = disk->status;
1253 info->disk.state = s & CONFIGURED_DISK ? (1 << MD_DISK_ACTIVE) : 0;
1254 /* we don't change info->disk.raid_disk here because
1255 * this state will be finalized in mdmon after we have
1256 * found the 'most fresh' version of the metadata
1257 */
1258 info->disk.state |= s & FAILED_DISK ? (1 << MD_DISK_FAULTY) : 0;
1259 info->disk.state |= s & SPARE_DISK ? 0 : (1 << MD_DISK_SYNC);
1260 }
1261
1262 /* only call uuid_from_super_imsm when this disk is part of a populated container,
1263 * ->compare_super may have updated the 'num_raid_devs' field for spares
1264 */
1265 if (info->disk.state & (1 << MD_DISK_SYNC) || super->anchor->num_raid_devs)
1266 uuid_from_super_imsm(st, info->uuid);
1267 else {
1268 memcpy(info->uuid, uuid_match_any, sizeof(int[4]));
1269 fixup_container_spare_uuid(info);
1270 }
1271 }
1272
1273 static int update_super_imsm(struct supertype *st, struct mdinfo *info,
1274 char *update, char *devname, int verbose,
1275 int uuid_set, char *homehost)
1276 {
1277 /* FIXME */
1278
1279 /* For 'assemble' and 'force' we need to return non-zero if any
1280 * change was made. For others, the return value is ignored.
1281 * Update options are:
1282 * force-one : This device looks a bit old but needs to be included,
1283 * update age info appropriately.
1284 * assemble: clear any 'faulty' flag to allow this device to
1285 * be assembled.
1286 * force-array: Array is degraded but being forced, mark it clean
1287 * if that will be needed to assemble it.
1288 *
1289 * newdev: not used ????
1290 * grow: Array has gained a new device - this is currently for
1291 * linear only
1292 * resync: mark as dirty so a resync will happen.
1293 * name: update the name - preserving the homehost
1294 *
1295 * Following are not relevant for this imsm:
1296 * sparc2.2 : update from old dodgey metadata
1297 * super-minor: change the preferred_minor number
1298 * summaries: update redundant counters.
1299 * uuid: Change the uuid of the array to match watch is given
1300 * homehost: update the recorded homehost
1301 * _reshape_progress: record new reshape_progress position.
1302 */
1303 int rv = 0;
1304 //struct intel_super *super = st->sb;
1305 //struct imsm_super *mpb = super->mpb;
1306
1307 if (strcmp(update, "grow") == 0) {
1308 }
1309 if (strcmp(update, "resync") == 0) {
1310 /* dev->vol.dirty = 1; */
1311 }
1312
1313 /* IMSM has no concept of UUID or homehost */
1314
1315 return rv;
1316 }
1317
1318 static size_t disks_to_mpb_size(int disks)
1319 {
1320 size_t size;
1321
1322 size = sizeof(struct imsm_super);
1323 size += (disks - 1) * sizeof(struct imsm_disk);
1324 size += 2 * sizeof(struct imsm_dev);
1325 /* up to 2 maps per raid device (-2 for imsm_maps in imsm_dev */
1326 size += (4 - 2) * sizeof(struct imsm_map);
1327 /* 4 possible disk_ord_tbl's */
1328 size += 4 * (disks - 1) * sizeof(__u32);
1329
1330 return size;
1331 }
1332
1333 static __u64 avail_size_imsm(struct supertype *st, __u64 devsize)
1334 {
1335 if (devsize < (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS))
1336 return 0;
1337
1338 return devsize - (MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS);
1339 }
1340
1341 static void free_devlist(struct intel_super *super)
1342 {
1343 struct intel_dev *dv;
1344
1345 while (super->devlist) {
1346 dv = super->devlist->next;
1347 free(super->devlist->dev);
1348 free(super->devlist);
1349 super->devlist = dv;
1350 }
1351 }
1352
1353 static void imsm_copy_dev(struct imsm_dev *dest, struct imsm_dev *src)
1354 {
1355 memcpy(dest, src, sizeof_imsm_dev(src, 0));
1356 }
1357
1358 static int compare_super_imsm(struct supertype *st, struct supertype *tst)
1359 {
1360 /*
1361 * return:
1362 * 0 same, or first was empty, and second was copied
1363 * 1 second had wrong number
1364 * 2 wrong uuid
1365 * 3 wrong other info
1366 */
1367 struct intel_super *first = st->sb;
1368 struct intel_super *sec = tst->sb;
1369
1370 if (!first) {
1371 st->sb = tst->sb;
1372 tst->sb = NULL;
1373 return 0;
1374 }
1375
1376 if (memcmp(first->anchor->sig, sec->anchor->sig, MAX_SIGNATURE_LENGTH) != 0)
1377 return 3;
1378
1379 /* if an anchor does not have num_raid_devs set then it is a free
1380 * floating spare
1381 */
1382 if (first->anchor->num_raid_devs > 0 &&
1383 sec->anchor->num_raid_devs > 0) {
1384 if (first->anchor->family_num != sec->anchor->family_num)
1385 return 3;
1386 }
1387
1388 /* if 'first' is a spare promote it to a populated mpb with sec's
1389 * family number
1390 */
1391 if (first->anchor->num_raid_devs == 0 &&
1392 sec->anchor->num_raid_devs > 0) {
1393 int i;
1394 struct intel_dev *dv;
1395 struct imsm_dev *dev;
1396
1397 /* we need to copy raid device info from sec if an allocation
1398 * fails here we don't associate the spare
1399 */
1400 for (i = 0; i < sec->anchor->num_raid_devs; i++) {
1401 dv = malloc(sizeof(*dv));
1402 if (!dv)
1403 break;
1404 dev = malloc(sizeof_imsm_dev(get_imsm_dev(sec, i), 1));
1405 if (!dev) {
1406 free(dv);
1407 break;
1408 }
1409 dv->dev = dev;
1410 dv->index = i;
1411 dv->next = first->devlist;
1412 first->devlist = dv;
1413 }
1414 if (i <= sec->anchor->num_raid_devs) {
1415 /* allocation failure */
1416 free_devlist(first);
1417 fprintf(stderr, "imsm: failed to associate spare\n");
1418 return 3;
1419 }
1420 for (i = 0; i < sec->anchor->num_raid_devs; i++)
1421 imsm_copy_dev(get_imsm_dev(first, i), get_imsm_dev(sec, i));
1422
1423 first->anchor->num_raid_devs = sec->anchor->num_raid_devs;
1424 first->anchor->family_num = sec->anchor->family_num;
1425 }
1426
1427 return 0;
1428 }
1429
1430 static void fd2devname(int fd, char *name)
1431 {
1432 struct stat st;
1433 char path[256];
1434 char dname[100];
1435 char *nm;
1436 int rv;
1437
1438 name[0] = '\0';
1439 if (fstat(fd, &st) != 0)
1440 return;
1441 sprintf(path, "/sys/dev/block/%d:%d",
1442 major(st.st_rdev), minor(st.st_rdev));
1443
1444 rv = readlink(path, dname, sizeof(dname));
1445 if (rv <= 0)
1446 return;
1447
1448 dname[rv] = '\0';
1449 nm = strrchr(dname, '/');
1450 nm++;
1451 snprintf(name, MAX_RAID_SERIAL_LEN, "/dev/%s", nm);
1452 }
1453
1454
1455 extern int scsi_get_serial(int fd, void *buf, size_t buf_len);
1456
1457 static int imsm_read_serial(int fd, char *devname,
1458 __u8 serial[MAX_RAID_SERIAL_LEN])
1459 {
1460 unsigned char scsi_serial[255];
1461 int rv;
1462 int rsp_len;
1463 int len;
1464 char *dest;
1465 char *src;
1466 char *rsp_buf;
1467 int i;
1468
1469 memset(scsi_serial, 0, sizeof(scsi_serial));
1470
1471 rv = scsi_get_serial(fd, scsi_serial, sizeof(scsi_serial));
1472
1473 if (rv && check_env("IMSM_DEVNAME_AS_SERIAL")) {
1474 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1475 fd2devname(fd, (char *) serial);
1476 return 0;
1477 }
1478
1479 if (rv != 0) {
1480 if (devname)
1481 fprintf(stderr,
1482 Name ": Failed to retrieve serial for %s\n",
1483 devname);
1484 return rv;
1485 }
1486
1487 rsp_len = scsi_serial[3];
1488 if (!rsp_len) {
1489 if (devname)
1490 fprintf(stderr,
1491 Name ": Failed to retrieve serial for %s\n",
1492 devname);
1493 return 2;
1494 }
1495 rsp_buf = (char *) &scsi_serial[4];
1496
1497 /* trim all whitespace and non-printable characters and convert
1498 * ':' to ';'
1499 */
1500 for (i = 0, dest = rsp_buf; i < rsp_len; i++) {
1501 src = &rsp_buf[i];
1502 if (*src > 0x20) {
1503 /* ':' is reserved for use in placeholder serial
1504 * numbers for missing disks
1505 */
1506 if (*src == ':')
1507 *dest++ = ';';
1508 else
1509 *dest++ = *src;
1510 }
1511 }
1512 len = dest - rsp_buf;
1513 dest = rsp_buf;
1514
1515 /* truncate leading characters */
1516 if (len > MAX_RAID_SERIAL_LEN) {
1517 dest += len - MAX_RAID_SERIAL_LEN;
1518 len = MAX_RAID_SERIAL_LEN;
1519 }
1520
1521 memset(serial, 0, MAX_RAID_SERIAL_LEN);
1522 memcpy(serial, dest, len);
1523
1524 return 0;
1525 }
1526
1527 static int serialcmp(__u8 *s1, __u8 *s2)
1528 {
1529 return strncmp((char *) s1, (char *) s2, MAX_RAID_SERIAL_LEN);
1530 }
1531
1532 static void serialcpy(__u8 *dest, __u8 *src)
1533 {
1534 strncpy((char *) dest, (char *) src, MAX_RAID_SERIAL_LEN);
1535 }
1536
1537 static struct dl *serial_to_dl(__u8 *serial, struct intel_super *super)
1538 {
1539 struct dl *dl;
1540
1541 for (dl = super->disks; dl; dl = dl->next)
1542 if (serialcmp(dl->serial, serial) == 0)
1543 break;
1544
1545 return dl;
1546 }
1547
1548 static int
1549 load_imsm_disk(int fd, struct intel_super *super, char *devname, int keep_fd)
1550 {
1551 struct dl *dl;
1552 struct stat stb;
1553 int rv;
1554 int i;
1555 int alloc = 1;
1556 __u8 serial[MAX_RAID_SERIAL_LEN];
1557
1558 rv = imsm_read_serial(fd, devname, serial);
1559
1560 if (rv != 0)
1561 return 2;
1562
1563 /* check if this is a disk we have seen before. it may be a spare in
1564 * super->disks while the current anchor believes it is a raid member,
1565 * check if we need to update dl->index
1566 */
1567 dl = serial_to_dl(serial, super);
1568 if (!dl)
1569 dl = malloc(sizeof(*dl));
1570 else
1571 alloc = 0;
1572
1573 if (!dl) {
1574 if (devname)
1575 fprintf(stderr,
1576 Name ": failed to allocate disk buffer for %s\n",
1577 devname);
1578 return 2;
1579 }
1580
1581 if (alloc) {
1582 fstat(fd, &stb);
1583 dl->major = major(stb.st_rdev);
1584 dl->minor = minor(stb.st_rdev);
1585 dl->next = super->disks;
1586 dl->fd = keep_fd ? fd : -1;
1587 dl->devname = devname ? strdup(devname) : NULL;
1588 serialcpy(dl->serial, serial);
1589 dl->index = -2;
1590 dl->e = NULL;
1591 } else if (keep_fd) {
1592 close(dl->fd);
1593 dl->fd = fd;
1594 }
1595
1596 /* look up this disk's index in the current anchor */
1597 for (i = 0; i < super->anchor->num_disks; i++) {
1598 struct imsm_disk *disk_iter;
1599
1600 disk_iter = __get_imsm_disk(super->anchor, i);
1601
1602 if (serialcmp(disk_iter->serial, dl->serial) == 0) {
1603 dl->disk = *disk_iter;
1604 /* only set index on disks that are a member of a
1605 * populated contianer, i.e. one with raid_devs
1606 */
1607 if (dl->disk.status & FAILED_DISK)
1608 dl->index = -2;
1609 else if (dl->disk.status & SPARE_DISK)
1610 dl->index = -1;
1611 else
1612 dl->index = i;
1613
1614 break;
1615 }
1616 }
1617
1618 /* no match, maybe a stale failed drive */
1619 if (i == super->anchor->num_disks && dl->index >= 0) {
1620 dl->disk = *__get_imsm_disk(super->anchor, dl->index);
1621 if (dl->disk.status & FAILED_DISK)
1622 dl->index = -2;
1623 }
1624
1625 if (alloc)
1626 super->disks = dl;
1627
1628 return 0;
1629 }
1630
1631 #ifndef MDASSEMBLE
1632 /* When migrating map0 contains the 'destination' state while map1
1633 * contains the current state. When not migrating map0 contains the
1634 * current state. This routine assumes that map[0].map_state is set to
1635 * the current array state before being called.
1636 *
1637 * Migration is indicated by one of the following states
1638 * 1/ Idle (migr_state=0 map0state=normal||unitialized||degraded||failed)
1639 * 2/ Initialize (migr_state=1 migr_type=MIGR_INIT map0state=normal
1640 * map1state=unitialized)
1641 * 3/ Verify (Resync) (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1642 * map1state=normal)
1643 * 4/ Rebuild (migr_state=1 migr_type=MIGR_REBUILD map0state=normal
1644 * map1state=degraded)
1645 */
1646 static void migrate(struct imsm_dev *dev, __u8 to_state, int migr_type)
1647 {
1648 struct imsm_map *dest;
1649 struct imsm_map *src = get_imsm_map(dev, 0);
1650
1651 dev->vol.migr_state = 1;
1652 dev->vol.migr_type = migr_type;
1653 dev->vol.curr_migr_unit = 0;
1654 dest = get_imsm_map(dev, 1);
1655
1656 /* duplicate and then set the target end state in map[0] */
1657 memcpy(dest, src, sizeof_imsm_map(src));
1658 if (migr_type == MIGR_REBUILD) {
1659 __u32 ord;
1660 int i;
1661
1662 for (i = 0; i < src->num_members; i++) {
1663 ord = __le32_to_cpu(src->disk_ord_tbl[i]);
1664 set_imsm_ord_tbl_ent(src, i, ord_to_idx(ord));
1665 }
1666 }
1667
1668 src->map_state = to_state;
1669 }
1670
1671 static void end_migration(struct imsm_dev *dev, __u8 map_state)
1672 {
1673 struct imsm_map *map = get_imsm_map(dev, 0);
1674 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
1675 int i;
1676
1677 /* merge any IMSM_ORD_REBUILD bits that were not successfully
1678 * completed in the last migration.
1679 *
1680 * FIXME add support for online capacity expansion and
1681 * raid-level-migration
1682 */
1683 for (i = 0; i < prev->num_members; i++)
1684 map->disk_ord_tbl[i] |= prev->disk_ord_tbl[i];
1685
1686 dev->vol.migr_state = 0;
1687 dev->vol.curr_migr_unit = 0;
1688 map->map_state = map_state;
1689 }
1690 #endif
1691
1692 static int parse_raid_devices(struct intel_super *super)
1693 {
1694 int i;
1695 struct imsm_dev *dev_new;
1696 size_t len, len_migr;
1697 size_t space_needed = 0;
1698 struct imsm_super *mpb = super->anchor;
1699
1700 for (i = 0; i < super->anchor->num_raid_devs; i++) {
1701 struct imsm_dev *dev_iter = __get_imsm_dev(super->anchor, i);
1702 struct intel_dev *dv;
1703
1704 len = sizeof_imsm_dev(dev_iter, 0);
1705 len_migr = sizeof_imsm_dev(dev_iter, 1);
1706 if (len_migr > len)
1707 space_needed += len_migr - len;
1708
1709 dv = malloc(sizeof(*dv));
1710 if (!dv)
1711 return 1;
1712 dev_new = malloc(len_migr);
1713 if (!dev_new) {
1714 free(dv);
1715 return 1;
1716 }
1717 imsm_copy_dev(dev_new, dev_iter);
1718 dv->dev = dev_new;
1719 dv->index = i;
1720 dv->next = super->devlist;
1721 super->devlist = dv;
1722 }
1723
1724 /* ensure that super->buf is large enough when all raid devices
1725 * are migrating
1726 */
1727 if (__le32_to_cpu(mpb->mpb_size) + space_needed > super->len) {
1728 void *buf;
1729
1730 len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + space_needed, 512);
1731 if (posix_memalign(&buf, 512, len) != 0)
1732 return 1;
1733
1734 memcpy(buf, super->buf, len);
1735 free(super->buf);
1736 super->buf = buf;
1737 super->len = len;
1738 }
1739
1740 return 0;
1741 }
1742
1743 /* retrieve a pointer to the bbm log which starts after all raid devices */
1744 struct bbm_log *__get_imsm_bbm_log(struct imsm_super *mpb)
1745 {
1746 void *ptr = NULL;
1747
1748 if (__le32_to_cpu(mpb->bbm_log_size)) {
1749 ptr = mpb;
1750 ptr += mpb->mpb_size - __le32_to_cpu(mpb->bbm_log_size);
1751 }
1752
1753 return ptr;
1754 }
1755
1756 static void __free_imsm(struct intel_super *super, int free_disks);
1757
1758 /* load_imsm_mpb - read matrix metadata
1759 * allocates super->mpb to be freed by free_super
1760 */
1761 static int load_imsm_mpb(int fd, struct intel_super *super, char *devname)
1762 {
1763 unsigned long long dsize;
1764 unsigned long long sectors;
1765 struct stat;
1766 struct imsm_super *anchor;
1767 __u32 check_sum;
1768 int rc;
1769
1770 get_dev_size(fd, NULL, &dsize);
1771
1772 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0) {
1773 if (devname)
1774 fprintf(stderr,
1775 Name ": Cannot seek to anchor block on %s: %s\n",
1776 devname, strerror(errno));
1777 return 1;
1778 }
1779
1780 if (posix_memalign((void**)&anchor, 512, 512) != 0) {
1781 if (devname)
1782 fprintf(stderr,
1783 Name ": Failed to allocate imsm anchor buffer"
1784 " on %s\n", devname);
1785 return 1;
1786 }
1787 if (read(fd, anchor, 512) != 512) {
1788 if (devname)
1789 fprintf(stderr,
1790 Name ": Cannot read anchor block on %s: %s\n",
1791 devname, strerror(errno));
1792 free(anchor);
1793 return 1;
1794 }
1795
1796 if (strncmp((char *) anchor->sig, MPB_SIGNATURE, MPB_SIG_LEN) != 0) {
1797 if (devname)
1798 fprintf(stderr,
1799 Name ": no IMSM anchor on %s\n", devname);
1800 free(anchor);
1801 return 2;
1802 }
1803
1804 __free_imsm(super, 0);
1805 super->len = ROUND_UP(anchor->mpb_size, 512);
1806 if (posix_memalign(&super->buf, 512, super->len) != 0) {
1807 if (devname)
1808 fprintf(stderr,
1809 Name ": unable to allocate %zu byte mpb buffer\n",
1810 super->len);
1811 free(anchor);
1812 return 2;
1813 }
1814 memcpy(super->buf, anchor, 512);
1815
1816 sectors = mpb_sectors(anchor) - 1;
1817 free(anchor);
1818 if (!sectors) {
1819 check_sum = __gen_imsm_checksum(super->anchor);
1820 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1821 if (devname)
1822 fprintf(stderr,
1823 Name ": IMSM checksum %x != %x on %s\n",
1824 check_sum,
1825 __le32_to_cpu(super->anchor->check_sum),
1826 devname);
1827 return 2;
1828 }
1829
1830 rc = load_imsm_disk(fd, super, devname, 0);
1831 if (rc == 0)
1832 rc = parse_raid_devices(super);
1833 return rc;
1834 }
1835
1836 /* read the extended mpb */
1837 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0) {
1838 if (devname)
1839 fprintf(stderr,
1840 Name ": Cannot seek to extended mpb on %s: %s\n",
1841 devname, strerror(errno));
1842 return 1;
1843 }
1844
1845 if (read(fd, super->buf + 512, super->len - 512) != super->len - 512) {
1846 if (devname)
1847 fprintf(stderr,
1848 Name ": Cannot read extended mpb on %s: %s\n",
1849 devname, strerror(errno));
1850 return 2;
1851 }
1852
1853 check_sum = __gen_imsm_checksum(super->anchor);
1854 if (check_sum != __le32_to_cpu(super->anchor->check_sum)) {
1855 if (devname)
1856 fprintf(stderr,
1857 Name ": IMSM checksum %x != %x on %s\n",
1858 check_sum, __le32_to_cpu(super->anchor->check_sum),
1859 devname);
1860 return 3;
1861 }
1862
1863 /* FIXME the BBM log is disk specific so we cannot use this global
1864 * buffer for all disks. Ok for now since we only look at the global
1865 * bbm_log_size parameter to gate assembly
1866 */
1867 super->bbm_log = __get_imsm_bbm_log(super->anchor);
1868
1869 rc = load_imsm_disk(fd, super, devname, 0);
1870 if (rc == 0)
1871 rc = parse_raid_devices(super);
1872
1873 return rc;
1874 }
1875
1876 static void __free_imsm_disk(struct dl *d)
1877 {
1878 if (d->fd >= 0)
1879 close(d->fd);
1880 if (d->devname)
1881 free(d->devname);
1882 if (d->e)
1883 free(d->e);
1884 free(d);
1885
1886 }
1887 static void free_imsm_disks(struct intel_super *super)
1888 {
1889 struct dl *d;
1890
1891 while (super->disks) {
1892 d = super->disks;
1893 super->disks = d->next;
1894 __free_imsm_disk(d);
1895 }
1896 while (super->missing) {
1897 d = super->missing;
1898 super->missing = d->next;
1899 __free_imsm_disk(d);
1900 }
1901
1902 }
1903
1904 /* free all the pieces hanging off of a super pointer */
1905 static void __free_imsm(struct intel_super *super, int free_disks)
1906 {
1907 if (super->buf) {
1908 free(super->buf);
1909 super->buf = NULL;
1910 }
1911 if (free_disks)
1912 free_imsm_disks(super);
1913 free_devlist(super);
1914 if (super->hba) {
1915 free((void *) super->hba);
1916 super->hba = NULL;
1917 }
1918 }
1919
1920 static void free_imsm(struct intel_super *super)
1921 {
1922 __free_imsm(super, 1);
1923 free(super);
1924 }
1925
1926 static void free_super_imsm(struct supertype *st)
1927 {
1928 struct intel_super *super = st->sb;
1929
1930 if (!super)
1931 return;
1932
1933 free_imsm(super);
1934 st->sb = NULL;
1935 }
1936
1937 static struct intel_super *alloc_super(int creating_imsm)
1938 {
1939 struct intel_super *super = malloc(sizeof(*super));
1940
1941 if (super) {
1942 memset(super, 0, sizeof(*super));
1943 super->creating_imsm = creating_imsm;
1944 super->current_vol = -1;
1945 super->create_offset = ~((__u32 ) 0);
1946 if (!check_env("IMSM_NO_PLATFORM"))
1947 super->orom = find_imsm_orom();
1948 if (super->orom && !check_env("IMSM_TEST_OROM")) {
1949 struct sys_dev *list, *ent;
1950
1951 /* find the first intel ahci controller */
1952 list = find_driver_devices("pci", "ahci");
1953 for (ent = list; ent; ent = ent->next)
1954 if (devpath_to_vendor(ent->path) == 0x8086)
1955 break;
1956 if (ent) {
1957 super->hba = ent->path;
1958 ent->path = NULL;
1959 }
1960 free_sys_dev(&list);
1961 }
1962 }
1963
1964 return super;
1965 }
1966
1967 #ifndef MDASSEMBLE
1968 /* find_missing - helper routine for load_super_imsm_all that identifies
1969 * disks that have disappeared from the system. This routine relies on
1970 * the mpb being uptodate, which it is at load time.
1971 */
1972 static int find_missing(struct intel_super *super)
1973 {
1974 int i;
1975 struct imsm_super *mpb = super->anchor;
1976 struct dl *dl;
1977 struct imsm_disk *disk;
1978
1979 for (i = 0; i < mpb->num_disks; i++) {
1980 disk = __get_imsm_disk(mpb, i);
1981 dl = serial_to_dl(disk->serial, super);
1982 if (dl)
1983 continue;
1984
1985 dl = malloc(sizeof(*dl));
1986 if (!dl)
1987 return 1;
1988 dl->major = 0;
1989 dl->minor = 0;
1990 dl->fd = -1;
1991 dl->devname = strdup("missing");
1992 dl->index = i;
1993 serialcpy(dl->serial, disk->serial);
1994 dl->disk = *disk;
1995 dl->e = NULL;
1996 dl->next = super->missing;
1997 super->missing = dl;
1998 }
1999
2000 return 0;
2001 }
2002
2003 static int load_super_imsm_all(struct supertype *st, int fd, void **sbp,
2004 char *devname, int keep_fd)
2005 {
2006 struct mdinfo *sra;
2007 struct intel_super *super;
2008 struct mdinfo *sd, *best = NULL;
2009 __u32 bestgen = 0;
2010 __u32 gen;
2011 char nm[20];
2012 int dfd;
2013 int rv;
2014 int devnum = fd2devnum(fd);
2015 int retry;
2016 enum sysfs_read_flags flags;
2017
2018 flags = GET_LEVEL|GET_VERSION|GET_DEVS|GET_STATE;
2019 if (mdmon_running(devnum))
2020 flags |= SKIP_GONE_DEVS;
2021
2022 /* check if 'fd' an opened container */
2023 sra = sysfs_read(fd, 0, flags);
2024 if (!sra)
2025 return 1;
2026
2027 if (sra->array.major_version != -1 ||
2028 sra->array.minor_version != -2 ||
2029 strcmp(sra->text_version, "imsm") != 0)
2030 return 1;
2031
2032 super = alloc_super(0);
2033 if (!super)
2034 return 1;
2035
2036 /* find the most up to date disk in this array, skipping spares */
2037 for (sd = sra->devs; sd; sd = sd->next) {
2038 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2039 dfd = dev_open(nm, keep_fd ? O_RDWR : O_RDONLY);
2040 if (dfd < 0) {
2041 free_imsm(super);
2042 return 2;
2043 }
2044 rv = load_imsm_mpb(dfd, super, NULL);
2045
2046 /* retry the load if we might have raced against mdmon */
2047 if (rv == 3 && mdmon_running(devnum))
2048 for (retry = 0; retry < 3; retry++) {
2049 usleep(3000);
2050 rv = load_imsm_mpb(dfd, super, NULL);
2051 if (rv != 3)
2052 break;
2053 }
2054 if (!keep_fd)
2055 close(dfd);
2056 if (rv == 0) {
2057 if (super->anchor->num_raid_devs == 0)
2058 gen = 0;
2059 else
2060 gen = __le32_to_cpu(super->anchor->generation_num);
2061 if (!best || gen > bestgen) {
2062 bestgen = gen;
2063 best = sd;
2064 }
2065 } else {
2066 free_imsm(super);
2067 return rv;
2068 }
2069 }
2070
2071 if (!best) {
2072 free_imsm(super);
2073 return 1;
2074 }
2075
2076 /* load the most up to date anchor */
2077 sprintf(nm, "%d:%d", best->disk.major, best->disk.minor);
2078 dfd = dev_open(nm, O_RDONLY);
2079 if (dfd < 0) {
2080 free_imsm(super);
2081 return 1;
2082 }
2083 rv = load_imsm_mpb(dfd, super, NULL);
2084 close(dfd);
2085 if (rv != 0) {
2086 free_imsm(super);
2087 return 2;
2088 }
2089
2090 /* re-parse the disk list with the current anchor */
2091 for (sd = sra->devs ; sd ; sd = sd->next) {
2092 sprintf(nm, "%d:%d", sd->disk.major, sd->disk.minor);
2093 dfd = dev_open(nm, keep_fd? O_RDWR : O_RDONLY);
2094 if (dfd < 0) {
2095 free_imsm(super);
2096 return 2;
2097 }
2098 load_imsm_disk(dfd, super, NULL, keep_fd);
2099 if (!keep_fd)
2100 close(dfd);
2101 }
2102
2103
2104 if (find_missing(super) != 0) {
2105 free_imsm(super);
2106 return 2;
2107 }
2108
2109 if (st->subarray[0]) {
2110 if (atoi(st->subarray) <= super->anchor->num_raid_devs)
2111 super->current_vol = atoi(st->subarray);
2112 else
2113 return 1;
2114 }
2115
2116 *sbp = super;
2117 st->container_dev = devnum;
2118 if (st->ss == NULL) {
2119 st->ss = &super_imsm;
2120 st->minor_version = 0;
2121 st->max_devs = IMSM_MAX_DEVICES;
2122 }
2123 st->loaded_container = 1;
2124
2125 return 0;
2126 }
2127 #endif
2128
2129 static int load_super_imsm(struct supertype *st, int fd, char *devname)
2130 {
2131 struct intel_super *super;
2132 int rv;
2133
2134 #ifndef MDASSEMBLE
2135 if (load_super_imsm_all(st, fd, &st->sb, devname, 1) == 0)
2136 return 0;
2137 #endif
2138 if (st->subarray[0])
2139 return 1; /* FIXME */
2140
2141 super = alloc_super(0);
2142 if (!super) {
2143 fprintf(stderr,
2144 Name ": malloc of %zu failed.\n",
2145 sizeof(*super));
2146 return 1;
2147 }
2148
2149 rv = load_imsm_mpb(fd, super, devname);
2150
2151 if (rv) {
2152 if (devname)
2153 fprintf(stderr,
2154 Name ": Failed to load all information "
2155 "sections on %s\n", devname);
2156 free_imsm(super);
2157 return rv;
2158 }
2159
2160 st->sb = super;
2161 if (st->ss == NULL) {
2162 st->ss = &super_imsm;
2163 st->minor_version = 0;
2164 st->max_devs = IMSM_MAX_DEVICES;
2165 }
2166 st->loaded_container = 0;
2167
2168 return 0;
2169 }
2170
2171 static __u16 info_to_blocks_per_strip(mdu_array_info_t *info)
2172 {
2173 if (info->level == 1)
2174 return 128;
2175 return info->chunk_size >> 9;
2176 }
2177
2178 static __u32 info_to_num_data_stripes(mdu_array_info_t *info)
2179 {
2180 __u32 num_stripes;
2181
2182 num_stripes = (info->size * 2) / info_to_blocks_per_strip(info);
2183 if (info->level == 1)
2184 num_stripes /= 2;
2185
2186 return num_stripes;
2187 }
2188
2189 static __u32 info_to_blocks_per_member(mdu_array_info_t *info)
2190 {
2191 if (info->level == 1)
2192 return info->size * 2;
2193 else
2194 return (info->size * 2) & ~(info_to_blocks_per_strip(info) - 1);
2195 }
2196
2197 static void imsm_update_version_info(struct intel_super *super)
2198 {
2199 /* update the version and attributes */
2200 struct imsm_super *mpb = super->anchor;
2201 char *version;
2202 struct imsm_dev *dev;
2203 struct imsm_map *map;
2204 int i;
2205
2206 for (i = 0; i < mpb->num_raid_devs; i++) {
2207 dev = get_imsm_dev(super, i);
2208 map = get_imsm_map(dev, 0);
2209 if (__le32_to_cpu(dev->size_high) > 0)
2210 mpb->attributes |= MPB_ATTRIB_2TB;
2211
2212 /* FIXME detect when an array spans a port multiplier */
2213 #if 0
2214 mpb->attributes |= MPB_ATTRIB_PM;
2215 #endif
2216
2217 if (mpb->num_raid_devs > 1 ||
2218 mpb->attributes != MPB_ATTRIB_CHECKSUM_VERIFY) {
2219 version = MPB_VERSION_ATTRIBS;
2220 switch (get_imsm_raid_level(map)) {
2221 case 0: mpb->attributes |= MPB_ATTRIB_RAID0; break;
2222 case 1: mpb->attributes |= MPB_ATTRIB_RAID1; break;
2223 case 10: mpb->attributes |= MPB_ATTRIB_RAID10; break;
2224 case 5: mpb->attributes |= MPB_ATTRIB_RAID5; break;
2225 }
2226 } else {
2227 if (map->num_members >= 5)
2228 version = MPB_VERSION_5OR6_DISK_ARRAY;
2229 else if (dev->status == DEV_CLONE_N_GO)
2230 version = MPB_VERSION_CNG;
2231 else if (get_imsm_raid_level(map) == 5)
2232 version = MPB_VERSION_RAID5;
2233 else if (map->num_members >= 3)
2234 version = MPB_VERSION_3OR4_DISK_ARRAY;
2235 else if (get_imsm_raid_level(map) == 1)
2236 version = MPB_VERSION_RAID1;
2237 else
2238 version = MPB_VERSION_RAID0;
2239 }
2240 strcpy(((char *) mpb->sig) + strlen(MPB_SIGNATURE), version);
2241 }
2242 }
2243
2244 static int init_super_imsm_volume(struct supertype *st, mdu_array_info_t *info,
2245 unsigned long long size, char *name,
2246 char *homehost, int *uuid)
2247 {
2248 /* We are creating a volume inside a pre-existing container.
2249 * so st->sb is already set.
2250 */
2251 struct intel_super *super = st->sb;
2252 struct imsm_super *mpb = super->anchor;
2253 struct intel_dev *dv;
2254 struct imsm_dev *dev;
2255 struct imsm_vol *vol;
2256 struct imsm_map *map;
2257 int idx = mpb->num_raid_devs;
2258 int i;
2259 unsigned long long array_blocks;
2260 size_t size_old, size_new;
2261
2262 if (super->orom && mpb->num_raid_devs >= super->orom->vpa) {
2263 fprintf(stderr, Name": This imsm-container already has the "
2264 "maximum of %d volumes\n", super->orom->vpa);
2265 return 0;
2266 }
2267
2268 /* ensure the mpb is large enough for the new data */
2269 size_old = __le32_to_cpu(mpb->mpb_size);
2270 size_new = disks_to_mpb_size(info->nr_disks);
2271 if (size_new > size_old) {
2272 void *mpb_new;
2273 size_t size_round = ROUND_UP(size_new, 512);
2274
2275 if (posix_memalign(&mpb_new, 512, size_round) != 0) {
2276 fprintf(stderr, Name": could not allocate new mpb\n");
2277 return 0;
2278 }
2279 memcpy(mpb_new, mpb, size_old);
2280 free(mpb);
2281 mpb = mpb_new;
2282 super->anchor = mpb_new;
2283 mpb->mpb_size = __cpu_to_le32(size_new);
2284 memset(mpb_new + size_old, 0, size_round - size_old);
2285 }
2286 super->current_vol = idx;
2287 /* when creating the first raid device in this container set num_disks
2288 * to zero, i.e. delete this spare and add raid member devices in
2289 * add_to_super_imsm_volume()
2290 */
2291 if (super->current_vol == 0)
2292 mpb->num_disks = 0;
2293
2294 for (i = 0; i < super->current_vol; i++) {
2295 dev = get_imsm_dev(super, i);
2296 if (strncmp((char *) dev->volume, name,
2297 MAX_RAID_SERIAL_LEN) == 0) {
2298 fprintf(stderr, Name": '%s' is already defined for this container\n",
2299 name);
2300 return 0;
2301 }
2302 }
2303
2304 sprintf(st->subarray, "%d", idx);
2305 dv = malloc(sizeof(*dv));
2306 if (!dv) {
2307 fprintf(stderr, Name ": failed to allocate device list entry\n");
2308 return 0;
2309 }
2310 dev = malloc(sizeof(*dev) + sizeof(__u32) * (info->raid_disks - 1));
2311 if (!dev) {
2312 free(dv);
2313 fprintf(stderr, Name": could not allocate raid device\n");
2314 return 0;
2315 }
2316 strncpy((char *) dev->volume, name, MAX_RAID_SERIAL_LEN);
2317 if (info->level == 1)
2318 array_blocks = info_to_blocks_per_member(info);
2319 else
2320 array_blocks = calc_array_size(info->level, info->raid_disks,
2321 info->layout, info->chunk_size,
2322 info->size*2);
2323 dev->size_low = __cpu_to_le32((__u32) array_blocks);
2324 dev->size_high = __cpu_to_le32((__u32) (array_blocks >> 32));
2325 dev->status = __cpu_to_le32(0);
2326 dev->reserved_blocks = __cpu_to_le32(0);
2327 vol = &dev->vol;
2328 vol->migr_state = 0;
2329 vol->migr_type = MIGR_INIT;
2330 vol->dirty = 0;
2331 vol->curr_migr_unit = 0;
2332 map = get_imsm_map(dev, 0);
2333 map->pba_of_lba0 = __cpu_to_le32(super->create_offset);
2334 map->blocks_per_member = __cpu_to_le32(info_to_blocks_per_member(info));
2335 map->blocks_per_strip = __cpu_to_le16(info_to_blocks_per_strip(info));
2336 map->num_data_stripes = __cpu_to_le32(info_to_num_data_stripes(info));
2337 map->failed_disk_num = ~0;
2338 map->map_state = info->level ? IMSM_T_STATE_UNINITIALIZED :
2339 IMSM_T_STATE_NORMAL;
2340
2341 if (info->level == 1 && info->raid_disks > 2) {
2342 fprintf(stderr, Name": imsm does not support more than 2 disks"
2343 "in a raid1 volume\n");
2344 return 0;
2345 }
2346 if (info->level == 10) {
2347 map->raid_level = 1;
2348 map->num_domains = info->raid_disks / 2;
2349 } else {
2350 map->raid_level = info->level;
2351 map->num_domains = !!map->raid_level;
2352 }
2353
2354 map->num_members = info->raid_disks;
2355 for (i = 0; i < map->num_members; i++) {
2356 /* initialized in add_to_super */
2357 set_imsm_ord_tbl_ent(map, i, 0);
2358 }
2359 mpb->num_raid_devs++;
2360
2361 dv->dev = dev;
2362 dv->index = super->current_vol;
2363 dv->next = super->devlist;
2364 super->devlist = dv;
2365
2366 imsm_update_version_info(super);
2367
2368 return 1;
2369 }
2370
2371 static int init_super_imsm(struct supertype *st, mdu_array_info_t *info,
2372 unsigned long long size, char *name,
2373 char *homehost, int *uuid)
2374 {
2375 /* This is primarily called by Create when creating a new array.
2376 * We will then get add_to_super called for each component, and then
2377 * write_init_super called to write it out to each device.
2378 * For IMSM, Create can create on fresh devices or on a pre-existing
2379 * array.
2380 * To create on a pre-existing array a different method will be called.
2381 * This one is just for fresh drives.
2382 */
2383 struct intel_super *super;
2384 struct imsm_super *mpb;
2385 size_t mpb_size;
2386 char *version;
2387
2388 if (!info) {
2389 st->sb = NULL;
2390 return 0;
2391 }
2392 if (st->sb)
2393 return init_super_imsm_volume(st, info, size, name, homehost,
2394 uuid);
2395
2396 super = alloc_super(1);
2397 if (!super)
2398 return 0;
2399 mpb_size = disks_to_mpb_size(info->nr_disks);
2400 if (posix_memalign(&super->buf, 512, mpb_size) != 0) {
2401 free(super);
2402 return 0;
2403 }
2404 mpb = super->buf;
2405 memset(mpb, 0, mpb_size);
2406
2407 mpb->attributes = MPB_ATTRIB_CHECKSUM_VERIFY;
2408
2409 version = (char *) mpb->sig;
2410 strcpy(version, MPB_SIGNATURE);
2411 version += strlen(MPB_SIGNATURE);
2412 strcpy(version, MPB_VERSION_RAID0);
2413 mpb->mpb_size = mpb_size;
2414
2415 st->sb = super;
2416 return 1;
2417 }
2418
2419 #ifndef MDASSEMBLE
2420 static int add_to_super_imsm_volume(struct supertype *st, mdu_disk_info_t *dk,
2421 int fd, char *devname)
2422 {
2423 struct intel_super *super = st->sb;
2424 struct imsm_super *mpb = super->anchor;
2425 struct dl *dl;
2426 struct imsm_dev *dev;
2427 struct imsm_map *map;
2428
2429 dev = get_imsm_dev(super, super->current_vol);
2430 map = get_imsm_map(dev, 0);
2431
2432 if (! (dk->state & (1<<MD_DISK_SYNC))) {
2433 fprintf(stderr, Name ": %s: Cannot add spare devices to IMSM volume\n",
2434 devname);
2435 return 1;
2436 }
2437
2438 if (fd == -1) {
2439 /* we're doing autolayout so grab the pre-marked (in
2440 * validate_geometry) raid_disk
2441 */
2442 for (dl = super->disks; dl; dl = dl->next)
2443 if (dl->raiddisk == dk->raid_disk)
2444 break;
2445 } else {
2446 for (dl = super->disks; dl ; dl = dl->next)
2447 if (dl->major == dk->major &&
2448 dl->minor == dk->minor)
2449 break;
2450 }
2451
2452 if (!dl) {
2453 fprintf(stderr, Name ": %s is not a member of the same container\n", devname);
2454 return 1;
2455 }
2456
2457 /* add a pristine spare to the metadata */
2458 if (dl->index < 0) {
2459 dl->index = super->anchor->num_disks;
2460 super->anchor->num_disks++;
2461 }
2462 set_imsm_ord_tbl_ent(map, dk->number, dl->index);
2463 dl->disk.status = CONFIGURED_DISK | USABLE_DISK;
2464
2465 /* if we are creating the first raid device update the family number */
2466 if (super->current_vol == 0) {
2467 __u32 sum;
2468 struct imsm_dev *_dev = __get_imsm_dev(mpb, 0);
2469 struct imsm_disk *_disk = __get_imsm_disk(mpb, dl->index);
2470
2471 *_dev = *dev;
2472 *_disk = dl->disk;
2473 sum = __gen_imsm_checksum(mpb);
2474 mpb->family_num = __cpu_to_le32(sum);
2475 }
2476
2477 return 0;
2478 }
2479
2480 static int add_to_super_imsm(struct supertype *st, mdu_disk_info_t *dk,
2481 int fd, char *devname)
2482 {
2483 struct intel_super *super = st->sb;
2484 struct dl *dd;
2485 unsigned long long size;
2486 __u32 id;
2487 int rv;
2488 struct stat stb;
2489
2490 /* if we are on an RAID enabled platform check that the disk is
2491 * attached to the raid controller
2492 */
2493 if (super->hba && !disk_attached_to_hba(fd, super->hba)) {
2494 fprintf(stderr,
2495 Name ": %s is not attached to the raid controller: %s\n",
2496 devname ? : "disk", super->hba);
2497 return 1;
2498 }
2499
2500 if (super->current_vol >= 0)
2501 return add_to_super_imsm_volume(st, dk, fd, devname);
2502
2503 fstat(fd, &stb);
2504 dd = malloc(sizeof(*dd));
2505 if (!dd) {
2506 fprintf(stderr,
2507 Name ": malloc failed %s:%d.\n", __func__, __LINE__);
2508 return 1;
2509 }
2510 memset(dd, 0, sizeof(*dd));
2511 dd->major = major(stb.st_rdev);
2512 dd->minor = minor(stb.st_rdev);
2513 dd->index = -1;
2514 dd->devname = devname ? strdup(devname) : NULL;
2515 dd->fd = fd;
2516 dd->e = NULL;
2517 rv = imsm_read_serial(fd, devname, dd->serial);
2518 if (rv) {
2519 fprintf(stderr,
2520 Name ": failed to retrieve scsi serial, aborting\n");
2521 free(dd);
2522 abort();
2523 }
2524
2525 get_dev_size(fd, NULL, &size);
2526 size /= 512;
2527 serialcpy(dd->disk.serial, dd->serial);
2528 dd->disk.total_blocks = __cpu_to_le32(size);
2529 dd->disk.status = USABLE_DISK | SPARE_DISK;
2530 if (sysfs_disk_to_scsi_id(fd, &id) == 0)
2531 dd->disk.scsi_id = __cpu_to_le32(id);
2532 else
2533 dd->disk.scsi_id = __cpu_to_le32(0);
2534
2535 if (st->update_tail) {
2536 dd->next = super->add;
2537 super->add = dd;
2538 } else {
2539 dd->next = super->disks;
2540 super->disks = dd;
2541 }
2542
2543 return 0;
2544 }
2545
2546 static int store_imsm_mpb(int fd, struct intel_super *super);
2547
2548 /* spare records have their own family number and do not have any defined raid
2549 * devices
2550 */
2551 static int write_super_imsm_spares(struct intel_super *super, int doclose)
2552 {
2553 struct imsm_super mpb_save;
2554 struct imsm_super *mpb = super->anchor;
2555 __u32 sum;
2556 struct dl *d;
2557
2558 mpb_save = *mpb;
2559 mpb->num_raid_devs = 0;
2560 mpb->num_disks = 1;
2561 mpb->mpb_size = sizeof(struct imsm_super);
2562 mpb->generation_num = __cpu_to_le32(1UL);
2563
2564 for (d = super->disks; d; d = d->next) {
2565 if (d->index != -1)
2566 continue;
2567
2568 mpb->disk[0] = d->disk;
2569 sum = __gen_imsm_checksum(mpb);
2570 mpb->family_num = __cpu_to_le32(sum);
2571 sum = __gen_imsm_checksum(mpb);
2572 mpb->check_sum = __cpu_to_le32(sum);
2573
2574 if (store_imsm_mpb(d->fd, super)) {
2575 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2576 __func__, d->major, d->minor, strerror(errno));
2577 *mpb = mpb_save;
2578 return 1;
2579 }
2580 if (doclose) {
2581 close(d->fd);
2582 d->fd = -1;
2583 }
2584 }
2585
2586 *mpb = mpb_save;
2587 return 0;
2588 }
2589
2590 static int write_super_imsm(struct intel_super *super, int doclose)
2591 {
2592 struct imsm_super *mpb = super->anchor;
2593 struct dl *d;
2594 __u32 generation;
2595 __u32 sum;
2596 int spares = 0;
2597 int i;
2598 __u32 mpb_size = sizeof(struct imsm_super) - sizeof(struct imsm_disk);
2599
2600 /* 'generation' is incremented everytime the metadata is written */
2601 generation = __le32_to_cpu(mpb->generation_num);
2602 generation++;
2603 mpb->generation_num = __cpu_to_le32(generation);
2604
2605 mpb_size += sizeof(struct imsm_disk) * mpb->num_disks;
2606 for (d = super->disks; d; d = d->next) {
2607 if (d->index == -1)
2608 spares++;
2609 else
2610 mpb->disk[d->index] = d->disk;
2611 }
2612 for (d = super->missing; d; d = d->next)
2613 mpb->disk[d->index] = d->disk;
2614
2615 for (i = 0; i < mpb->num_raid_devs; i++) {
2616 struct imsm_dev *dev = __get_imsm_dev(mpb, i);
2617
2618 imsm_copy_dev(dev, get_imsm_dev(super, i));
2619 mpb_size += sizeof_imsm_dev(dev, 0);
2620 }
2621 mpb_size += __le32_to_cpu(mpb->bbm_log_size);
2622 mpb->mpb_size = __cpu_to_le32(mpb_size);
2623
2624 /* recalculate checksum */
2625 sum = __gen_imsm_checksum(mpb);
2626 mpb->check_sum = __cpu_to_le32(sum);
2627
2628 /* write the mpb for disks that compose raid devices */
2629 for (d = super->disks; d ; d = d->next) {
2630 if (d->index < 0)
2631 continue;
2632 if (store_imsm_mpb(d->fd, super))
2633 fprintf(stderr, "%s: failed for device %d:%d %s\n",
2634 __func__, d->major, d->minor, strerror(errno));
2635 if (doclose) {
2636 close(d->fd);
2637 d->fd = -1;
2638 }
2639 }
2640
2641 if (spares)
2642 return write_super_imsm_spares(super, doclose);
2643
2644 return 0;
2645 }
2646
2647
2648 static int create_array(struct supertype *st)
2649 {
2650 size_t len;
2651 struct imsm_update_create_array *u;
2652 struct intel_super *super = st->sb;
2653 struct imsm_dev *dev = get_imsm_dev(super, super->current_vol);
2654 struct imsm_map *map = get_imsm_map(dev, 0);
2655 struct disk_info *inf;
2656 struct imsm_disk *disk;
2657 int i;
2658 int idx;
2659
2660 len = sizeof(*u) - sizeof(*dev) + sizeof_imsm_dev(dev, 0) +
2661 sizeof(*inf) * map->num_members;
2662 u = malloc(len);
2663 if (!u) {
2664 fprintf(stderr, "%s: failed to allocate update buffer\n",
2665 __func__);
2666 return 1;
2667 }
2668
2669 u->type = update_create_array;
2670 u->dev_idx = super->current_vol;
2671 imsm_copy_dev(&u->dev, dev);
2672 inf = get_disk_info(u);
2673 for (i = 0; i < map->num_members; i++) {
2674 idx = get_imsm_disk_idx(dev, i);
2675 disk = get_imsm_disk(super, idx);
2676 serialcpy(inf[i].serial, disk->serial);
2677 }
2678 append_metadata_update(st, u, len);
2679
2680 return 0;
2681 }
2682
2683 static int _add_disk(struct supertype *st)
2684 {
2685 struct intel_super *super = st->sb;
2686 size_t len;
2687 struct imsm_update_add_disk *u;
2688
2689 if (!super->add)
2690 return 0;
2691
2692 len = sizeof(*u);
2693 u = malloc(len);
2694 if (!u) {
2695 fprintf(stderr, "%s: failed to allocate update buffer\n",
2696 __func__);
2697 return 1;
2698 }
2699
2700 u->type = update_add_disk;
2701 append_metadata_update(st, u, len);
2702
2703 return 0;
2704 }
2705
2706 static int write_init_super_imsm(struct supertype *st)
2707 {
2708 if (st->update_tail) {
2709 /* queue the recently created array / added disk
2710 * as a metadata update */
2711 struct intel_super *super = st->sb;
2712 struct dl *d;
2713 int rv;
2714
2715 /* determine if we are creating a volume or adding a disk */
2716 if (super->current_vol < 0) {
2717 /* in the add disk case we are running in mdmon
2718 * context, so don't close fd's
2719 */
2720 return _add_disk(st);
2721 } else
2722 rv = create_array(st);
2723
2724 for (d = super->disks; d ; d = d->next) {
2725 close(d->fd);
2726 d->fd = -1;
2727 }
2728
2729 return rv;
2730 } else
2731 return write_super_imsm(st->sb, 1);
2732 }
2733 #endif
2734
2735 static int store_zero_imsm(struct supertype *st, int fd)
2736 {
2737 unsigned long long dsize;
2738 void *buf;
2739
2740 get_dev_size(fd, NULL, &dsize);
2741
2742 /* first block is stored on second to last sector of the disk */
2743 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
2744 return 1;
2745
2746 if (posix_memalign(&buf, 512, 512) != 0)
2747 return 1;
2748
2749 memset(buf, 0, 512);
2750 if (write(fd, buf, 512) != 512)
2751 return 1;
2752 return 0;
2753 }
2754
2755 static int imsm_bbm_log_size(struct imsm_super *mpb)
2756 {
2757 return __le32_to_cpu(mpb->bbm_log_size);
2758 }
2759
2760 #ifndef MDASSEMBLE
2761 static int validate_geometry_imsm_container(struct supertype *st, int level,
2762 int layout, int raiddisks, int chunk,
2763 unsigned long long size, char *dev,
2764 unsigned long long *freesize,
2765 int verbose)
2766 {
2767 int fd;
2768 unsigned long long ldsize;
2769 const struct imsm_orom *orom;
2770
2771 if (level != LEVEL_CONTAINER)
2772 return 0;
2773 if (!dev)
2774 return 1;
2775
2776 if (check_env("IMSM_NO_PLATFORM"))
2777 orom = NULL;
2778 else
2779 orom = find_imsm_orom();
2780 if (orom && raiddisks > orom->tds) {
2781 if (verbose)
2782 fprintf(stderr, Name ": %d exceeds maximum number of"
2783 " platform supported disks: %d\n",
2784 raiddisks, orom->tds);
2785 return 0;
2786 }
2787
2788 fd = open(dev, O_RDONLY|O_EXCL, 0);
2789 if (fd < 0) {
2790 if (verbose)
2791 fprintf(stderr, Name ": imsm: Cannot open %s: %s\n",
2792 dev, strerror(errno));
2793 return 0;
2794 }
2795 if (!get_dev_size(fd, dev, &ldsize)) {
2796 close(fd);
2797 return 0;
2798 }
2799 close(fd);
2800
2801 *freesize = avail_size_imsm(st, ldsize >> 9);
2802
2803 return 1;
2804 }
2805
2806 static unsigned long long find_size(struct extent *e, int *idx, int num_extents)
2807 {
2808 const unsigned long long base_start = e[*idx].start;
2809 unsigned long long end = base_start + e[*idx].size;
2810 int i;
2811
2812 if (base_start == end)
2813 return 0;
2814
2815 *idx = *idx + 1;
2816 for (i = *idx; i < num_extents; i++) {
2817 /* extend overlapping extents */
2818 if (e[i].start >= base_start &&
2819 e[i].start <= end) {
2820 if (e[i].size == 0)
2821 return 0;
2822 if (e[i].start + e[i].size > end)
2823 end = e[i].start + e[i].size;
2824 } else if (e[i].start > end) {
2825 *idx = i;
2826 break;
2827 }
2828 }
2829
2830 return end - base_start;
2831 }
2832
2833 static unsigned long long merge_extents(struct intel_super *super, int sum_extents)
2834 {
2835 /* build a composite disk with all known extents and generate a new
2836 * 'maxsize' given the "all disks in an array must share a common start
2837 * offset" constraint
2838 */
2839 struct extent *e = calloc(sum_extents, sizeof(*e));
2840 struct dl *dl;
2841 int i, j;
2842 int start_extent;
2843 unsigned long long pos;
2844 unsigned long long start = 0;
2845 unsigned long long maxsize;
2846 unsigned long reserve;
2847
2848 if (!e)
2849 return ~0ULL; /* error */
2850
2851 /* coalesce and sort all extents. also, check to see if we need to
2852 * reserve space between member arrays
2853 */
2854 j = 0;
2855 for (dl = super->disks; dl; dl = dl->next) {
2856 if (!dl->e)
2857 continue;
2858 for (i = 0; i < dl->extent_cnt; i++)
2859 e[j++] = dl->e[i];
2860 }
2861 qsort(e, sum_extents, sizeof(*e), cmp_extent);
2862
2863 /* merge extents */
2864 i = 0;
2865 j = 0;
2866 while (i < sum_extents) {
2867 e[j].start = e[i].start;
2868 e[j].size = find_size(e, &i, sum_extents);
2869 j++;
2870 if (e[j-1].size == 0)
2871 break;
2872 }
2873
2874 pos = 0;
2875 maxsize = 0;
2876 start_extent = 0;
2877 i = 0;
2878 do {
2879 unsigned long long esize;
2880
2881 esize = e[i].start - pos;
2882 if (esize >= maxsize) {
2883 maxsize = esize;
2884 start = pos;
2885 start_extent = i;
2886 }
2887 pos = e[i].start + e[i].size;
2888 i++;
2889 } while (e[i-1].size);
2890 free(e);
2891
2892 if (start_extent > 0)
2893 reserve = IMSM_RESERVED_SECTORS; /* gap between raid regions */
2894 else
2895 reserve = 0;
2896
2897 if (maxsize < reserve)
2898 return ~0ULL;
2899
2900 super->create_offset = ~((__u32) 0);
2901 if (start + reserve > super->create_offset)
2902 return ~0ULL; /* start overflows create_offset */
2903 super->create_offset = start + reserve;
2904
2905 return maxsize - reserve;
2906 }
2907
2908 static int is_raid_level_supported(const struct imsm_orom *orom, int level, int raiddisks)
2909 {
2910 if (level < 0 || level == 6 || level == 4)
2911 return 0;
2912
2913 /* if we have an orom prevent invalid raid levels */
2914 if (orom)
2915 switch (level) {
2916 case 0: return imsm_orom_has_raid0(orom);
2917 case 1:
2918 if (raiddisks > 2)
2919 return imsm_orom_has_raid1e(orom);
2920 return imsm_orom_has_raid1(orom) && raiddisks == 2;
2921 case 10: return imsm_orom_has_raid10(orom) && raiddisks == 4;
2922 case 5: return imsm_orom_has_raid5(orom) && raiddisks > 2;
2923 }
2924 else
2925 return 1; /* not on an Intel RAID platform so anything goes */
2926
2927 return 0;
2928 }
2929
2930 #define pr_vrb(fmt, arg...) (void) (verbose && fprintf(stderr, Name fmt, ##arg))
2931 /* validate_geometry_imsm_volume - lifted from validate_geometry_ddf_bvd
2932 * FIX ME add ahci details
2933 */
2934 static int validate_geometry_imsm_volume(struct supertype *st, int level,
2935 int layout, int raiddisks, int chunk,
2936 unsigned long long size, char *dev,
2937 unsigned long long *freesize,
2938 int verbose)
2939 {
2940 struct stat stb;
2941 struct intel_super *super = st->sb;
2942 struct imsm_super *mpb = super->anchor;
2943 struct dl *dl;
2944 unsigned long long pos = 0;
2945 unsigned long long maxsize;
2946 struct extent *e;
2947 int i;
2948
2949 /* We must have the container info already read in. */
2950 if (!super)
2951 return 0;
2952
2953 if (!is_raid_level_supported(super->orom, level, raiddisks)) {
2954 pr_vrb(": platform does not support raid%d with %d disk%s\n",
2955 level, raiddisks, raiddisks > 1 ? "s" : "");
2956 return 0;
2957 }
2958 if (super->orom && level != 1 &&
2959 !imsm_orom_has_chunk(super->orom, chunk)) {
2960 pr_vrb(": platform does not support a chunk size of: %d\n", chunk);
2961 return 0;
2962 }
2963 if (layout != imsm_level_to_layout(level)) {
2964 if (level == 5)
2965 pr_vrb(": imsm raid 5 only supports the left-asymmetric layout\n");
2966 else if (level == 10)
2967 pr_vrb(": imsm raid 10 only supports the n2 layout\n");
2968 else
2969 pr_vrb(": imsm unknown layout %#x for this raid level %d\n",
2970 layout, level);
2971 return 0;
2972 }
2973
2974 if (!dev) {
2975 /* General test: make sure there is space for
2976 * 'raiddisks' device extents of size 'size' at a given
2977 * offset
2978 */
2979 unsigned long long minsize = size;
2980 unsigned long long start_offset = ~0ULL;
2981 int dcnt = 0;
2982 if (minsize == 0)
2983 minsize = MPB_SECTOR_CNT + IMSM_RESERVED_SECTORS;
2984 for (dl = super->disks; dl ; dl = dl->next) {
2985 int found = 0;
2986
2987 pos = 0;
2988 i = 0;
2989 e = get_extents(super, dl);
2990 if (!e) continue;
2991 do {
2992 unsigned long long esize;
2993 esize = e[i].start - pos;
2994 if (esize >= minsize)
2995 found = 1;
2996 if (found && start_offset == ~0ULL) {
2997 start_offset = pos;
2998 break;
2999 } else if (found && pos != start_offset) {
3000 found = 0;
3001 break;
3002 }
3003 pos = e[i].start + e[i].size;
3004 i++;
3005 } while (e[i-1].size);
3006 if (found)
3007 dcnt++;
3008 free(e);
3009 }
3010 if (dcnt < raiddisks) {
3011 if (verbose)
3012 fprintf(stderr, Name ": imsm: Not enough "
3013 "devices with space for this array "
3014 "(%d < %d)\n",
3015 dcnt, raiddisks);
3016 return 0;
3017 }
3018 return 1;
3019 }
3020
3021 /* This device must be a member of the set */
3022 if (stat(dev, &stb) < 0)
3023 return 0;
3024 if ((S_IFMT & stb.st_mode) != S_IFBLK)
3025 return 0;
3026 for (dl = super->disks ; dl ; dl = dl->next) {
3027 if (dl->major == major(stb.st_rdev) &&
3028 dl->minor == minor(stb.st_rdev))
3029 break;
3030 }
3031 if (!dl) {
3032 if (verbose)
3033 fprintf(stderr, Name ": %s is not in the "
3034 "same imsm set\n", dev);
3035 return 0;
3036 } else if (super->orom && dl->index < 0 && mpb->num_raid_devs) {
3037 /* If a volume is present then the current creation attempt
3038 * cannot incorporate new spares because the orom may not
3039 * understand this configuration (all member disks must be
3040 * members of each array in the container).
3041 */
3042 fprintf(stderr, Name ": %s is a spare and a volume"
3043 " is already defined for this container\n", dev);
3044 fprintf(stderr, Name ": The option-rom requires all member"
3045 " disks to be a member of all volumes\n");
3046 return 0;
3047 }
3048
3049 /* retrieve the largest free space block */
3050 e = get_extents(super, dl);
3051 maxsize = 0;
3052 i = 0;
3053 if (e) {
3054 do {
3055 unsigned long long esize;
3056
3057 esize = e[i].start - pos;
3058 if (esize >= maxsize)
3059 maxsize = esize;
3060 pos = e[i].start + e[i].size;
3061 i++;
3062 } while (e[i-1].size);
3063 dl->e = e;
3064 dl->extent_cnt = i;
3065 } else {
3066 if (verbose)
3067 fprintf(stderr, Name ": unable to determine free space for: %s\n",
3068 dev);
3069 return 0;
3070 }
3071 if (maxsize < size) {
3072 if (verbose)
3073 fprintf(stderr, Name ": %s not enough space (%llu < %llu)\n",
3074 dev, maxsize, size);
3075 return 0;
3076 }
3077
3078 /* count total number of extents for merge */
3079 i = 0;
3080 for (dl = super->disks; dl; dl = dl->next)
3081 if (dl->e)
3082 i += dl->extent_cnt;
3083
3084 maxsize = merge_extents(super, i);
3085 if (maxsize < size) {
3086 if (verbose)
3087 fprintf(stderr, Name ": not enough space after merge (%llu < %llu)\n",
3088 maxsize, size);
3089 return 0;
3090 } else if (maxsize == ~0ULL) {
3091 if (verbose)
3092 fprintf(stderr, Name ": failed to merge %d extents\n", i);
3093 return 0;
3094 }
3095
3096 *freesize = maxsize;
3097
3098 return 1;
3099 }
3100
3101 static int reserve_space(struct supertype *st, int raiddisks,
3102 unsigned long long size, int chunk,
3103 unsigned long long *freesize)
3104 {
3105 struct intel_super *super = st->sb;
3106 struct imsm_super *mpb = super->anchor;
3107 struct dl *dl;
3108 int i;
3109 int extent_cnt;
3110 struct extent *e;
3111 unsigned long long maxsize;
3112 unsigned long long minsize;
3113 int cnt;
3114 int used;
3115
3116 /* find the largest common start free region of the possible disks */
3117 used = 0;
3118 extent_cnt = 0;
3119 cnt = 0;
3120 for (dl = super->disks; dl; dl = dl->next) {
3121 dl->raiddisk = -1;
3122
3123 if (dl->index >= 0)
3124 used++;
3125
3126 /* don't activate new spares if we are orom constrained
3127 * and there is already a volume active in the container
3128 */
3129 if (super->orom && dl->index < 0 && mpb->num_raid_devs)
3130 continue;
3131
3132 e = get_extents(super, dl);
3133 if (!e)
3134 continue;
3135 for (i = 1; e[i-1].size; i++)
3136 ;
3137 dl->e = e;
3138 dl->extent_cnt = i;
3139 extent_cnt += i;
3140 cnt++;
3141 }
3142
3143 maxsize = merge_extents(super, extent_cnt);
3144 minsize = size;
3145 if (size == 0)
3146 minsize = chunk;
3147
3148 if (cnt < raiddisks ||
3149 (super->orom && used && used != raiddisks) ||
3150 maxsize < minsize) {
3151 fprintf(stderr, Name ": not enough devices with space to create array.\n");
3152 return 0; /* No enough free spaces large enough */
3153 }
3154
3155 if (size == 0) {
3156 size = maxsize;
3157 if (chunk) {
3158 size /= chunk;
3159 size *= chunk;
3160 }
3161 }
3162
3163 cnt = 0;
3164 for (dl = super->disks; dl; dl = dl->next)
3165 if (dl->e)
3166 dl->raiddisk = cnt++;
3167
3168 *freesize = size;
3169
3170 return 1;
3171 }
3172
3173 static int validate_geometry_imsm(struct supertype *st, int level, int layout,
3174 int raiddisks, int chunk, unsigned long long size,
3175 char *dev, unsigned long long *freesize,
3176 int verbose)
3177 {
3178 int fd, cfd;
3179 struct mdinfo *sra;
3180
3181 /* if given unused devices create a container
3182 * if given given devices in a container create a member volume
3183 */
3184 if (level == LEVEL_CONTAINER) {
3185 /* Must be a fresh device to add to a container */
3186 return validate_geometry_imsm_container(st, level, layout,
3187 raiddisks, chunk, size,
3188 dev, freesize,
3189 verbose);
3190 }
3191
3192 if (!dev) {
3193 if (st->sb && freesize) {
3194 /* we are being asked to automatically layout a
3195 * new volume based on the current contents of
3196 * the container. If the the parameters can be
3197 * satisfied reserve_space will record the disks,
3198 * start offset, and size of the volume to be
3199 * created. add_to_super and getinfo_super
3200 * detect when autolayout is in progress.
3201 */
3202 return reserve_space(st, raiddisks, size, chunk, freesize);
3203 }
3204 return 1;
3205 }
3206 if (st->sb) {
3207 /* creating in a given container */
3208 return validate_geometry_imsm_volume(st, level, layout,
3209 raiddisks, chunk, size,
3210 dev, freesize, verbose);
3211 }
3212
3213 /* limit creation to the following levels */
3214 if (!dev)
3215 switch (level) {
3216 case 0:
3217 case 1:
3218 case 10:
3219 case 5:
3220 break;
3221 default:
3222 return 1;
3223 }
3224
3225 /* This device needs to be a device in an 'imsm' container */
3226 fd = open(dev, O_RDONLY|O_EXCL, 0);
3227 if (fd >= 0) {
3228 if (verbose)
3229 fprintf(stderr,
3230 Name ": Cannot create this array on device %s\n",
3231 dev);
3232 close(fd);
3233 return 0;
3234 }
3235 if (errno != EBUSY || (fd = open(dev, O_RDONLY, 0)) < 0) {
3236 if (verbose)
3237 fprintf(stderr, Name ": Cannot open %s: %s\n",
3238 dev, strerror(errno));
3239 return 0;
3240 }
3241 /* Well, it is in use by someone, maybe an 'imsm' container. */
3242 cfd = open_container(fd);
3243 if (cfd < 0) {
3244 close(fd);
3245 if (verbose)
3246 fprintf(stderr, Name ": Cannot use %s: It is busy\n",
3247 dev);
3248 return 0;
3249 }
3250 sra = sysfs_read(cfd, 0, GET_VERSION);
3251 close(fd);
3252 if (sra && sra->array.major_version == -1 &&
3253 strcmp(sra->text_version, "imsm") == 0) {
3254 /* This is a member of a imsm container. Load the container
3255 * and try to create a volume
3256 */
3257 struct intel_super *super;
3258
3259 if (load_super_imsm_all(st, cfd, (void **) &super, NULL, 1) == 0) {
3260 st->sb = super;
3261 st->container_dev = fd2devnum(cfd);
3262 close(cfd);
3263 return validate_geometry_imsm_volume(st, level, layout,
3264 raiddisks, chunk,
3265 size, dev,
3266 freesize, verbose);
3267 }
3268 close(cfd);
3269 } else /* may belong to another container */
3270 return 0;
3271
3272 return 1;
3273 }
3274 #endif /* MDASSEMBLE */
3275
3276 static struct mdinfo *container_content_imsm(struct supertype *st)
3277 {
3278 /* Given a container loaded by load_super_imsm_all,
3279 * extract information about all the arrays into
3280 * an mdinfo tree.
3281 *
3282 * For each imsm_dev create an mdinfo, fill it in,
3283 * then look for matching devices in super->disks
3284 * and create appropriate device mdinfo.
3285 */
3286 struct intel_super *super = st->sb;
3287 struct imsm_super *mpb = super->anchor;
3288 struct mdinfo *rest = NULL;
3289 int i;
3290
3291 /* do not assemble arrays that might have bad blocks */
3292 if (imsm_bbm_log_size(super->anchor)) {
3293 fprintf(stderr, Name ": BBM log found in metadata. "
3294 "Cannot activate array(s).\n");
3295 return NULL;
3296 }
3297
3298 for (i = 0; i < mpb->num_raid_devs; i++) {
3299 struct imsm_dev *dev = get_imsm_dev(super, i);
3300 struct imsm_map *map = get_imsm_map(dev, 0);
3301 struct mdinfo *this;
3302 int slot;
3303
3304 this = malloc(sizeof(*this));
3305 memset(this, 0, sizeof(*this));
3306 this->next = rest;
3307
3308 super->current_vol = i;
3309 getinfo_super_imsm_volume(st, this);
3310 for (slot = 0 ; slot < map->num_members; slot++) {
3311 struct mdinfo *info_d;
3312 struct dl *d;
3313 int idx;
3314 int skip;
3315 __u32 s;
3316 __u32 ord;
3317
3318 skip = 0;
3319 idx = get_imsm_disk_idx(dev, slot);
3320 ord = get_imsm_ord_tbl_ent(dev, slot);
3321 for (d = super->disks; d ; d = d->next)
3322 if (d->index == idx)
3323 break;
3324
3325 if (d == NULL)
3326 skip = 1;
3327
3328 s = d ? d->disk.status : 0;
3329 if (s & FAILED_DISK)
3330 skip = 1;
3331 if (!(s & USABLE_DISK))
3332 skip = 1;
3333 if (ord & IMSM_ORD_REBUILD)
3334 skip = 1;
3335
3336 /*
3337 * if we skip some disks the array will be assmebled degraded;
3338 * reset resync start to avoid a dirty-degraded situation
3339 *
3340 * FIXME handle dirty degraded
3341 */
3342 if (skip && !dev->vol.dirty)
3343 this->resync_start = ~0ULL;
3344 if (skip)
3345 continue;
3346
3347 info_d = malloc(sizeof(*info_d));
3348 if (!info_d) {
3349 fprintf(stderr, Name ": failed to allocate disk"
3350 " for volume %s\n", (char *) dev->volume);
3351 free(this);
3352 this = rest;
3353 break;
3354 }
3355 memset(info_d, 0, sizeof(*info_d));
3356 info_d->next = this->devs;
3357 this->devs = info_d;
3358
3359 info_d->disk.number = d->index;
3360 info_d->disk.major = d->major;
3361 info_d->disk.minor = d->minor;
3362 info_d->disk.raid_disk = slot;
3363
3364 this->array.working_disks++;
3365
3366 info_d->events = __le32_to_cpu(mpb->generation_num);
3367 info_d->data_offset = __le32_to_cpu(map->pba_of_lba0);
3368 info_d->component_size = __le32_to_cpu(map->blocks_per_member);
3369 if (d->devname)
3370 strcpy(info_d->name, d->devname);
3371 }
3372 rest = this;
3373 }
3374
3375 return rest;
3376 }
3377
3378
3379 #ifndef MDASSEMBLE
3380 static int imsm_open_new(struct supertype *c, struct active_array *a,
3381 char *inst)
3382 {
3383 struct intel_super *super = c->sb;
3384 struct imsm_super *mpb = super->anchor;
3385
3386 if (atoi(inst) >= mpb->num_raid_devs) {
3387 fprintf(stderr, "%s: subarry index %d, out of range\n",
3388 __func__, atoi(inst));
3389 return -ENODEV;
3390 }
3391
3392 dprintf("imsm: open_new %s\n", inst);
3393 a->info.container_member = atoi(inst);
3394 return 0;
3395 }
3396
3397 static __u8 imsm_check_degraded(struct intel_super *super, struct imsm_dev *dev, int failed)
3398 {
3399 struct imsm_map *map = get_imsm_map(dev, 0);
3400
3401 if (!failed)
3402 return map->map_state == IMSM_T_STATE_UNINITIALIZED ?
3403 IMSM_T_STATE_UNINITIALIZED : IMSM_T_STATE_NORMAL;
3404
3405 switch (get_imsm_raid_level(map)) {
3406 case 0:
3407 return IMSM_T_STATE_FAILED;
3408 break;
3409 case 1:
3410 if (failed < map->num_members)
3411 return IMSM_T_STATE_DEGRADED;
3412 else
3413 return IMSM_T_STATE_FAILED;
3414 break;
3415 case 10:
3416 {
3417 /**
3418 * check to see if any mirrors have failed, otherwise we
3419 * are degraded. Even numbered slots are mirrored on
3420 * slot+1
3421 */
3422 int i;
3423 /* gcc -Os complains that this is unused */
3424 int insync = insync;
3425
3426 for (i = 0; i < map->num_members; i++) {
3427 __u32 ord = get_imsm_ord_tbl_ent(dev, i);
3428 int idx = ord_to_idx(ord);
3429 struct imsm_disk *disk;
3430
3431 /* reset the potential in-sync count on even-numbered
3432 * slots. num_copies is always 2 for imsm raid10
3433 */
3434 if ((i & 1) == 0)
3435 insync = 2;
3436
3437 disk = get_imsm_disk(super, idx);
3438 if (!disk || disk->status & FAILED_DISK ||
3439 ord & IMSM_ORD_REBUILD)
3440 insync--;
3441
3442 /* no in-sync disks left in this mirror the
3443 * array has failed
3444 */
3445 if (insync == 0)
3446 return IMSM_T_STATE_FAILED;
3447 }
3448
3449 return IMSM_T_STATE_DEGRADED;
3450 }
3451 case 5:
3452 if (failed < 2)
3453 return IMSM_T_STATE_DEGRADED;
3454 else
3455 return IMSM_T_STATE_FAILED;
3456 break;
3457 default:
3458 break;
3459 }
3460
3461 return map->map_state;
3462 }
3463
3464 static int imsm_count_failed(struct intel_super *super, struct imsm_dev *dev)
3465 {
3466 int i;
3467 int failed = 0;
3468 struct imsm_disk *disk;
3469 struct imsm_map *map = get_imsm_map(dev, 0);
3470 struct imsm_map *prev = get_imsm_map(dev, dev->vol.migr_state);
3471 __u32 ord;
3472 int idx;
3473
3474 /* at the beginning of migration we set IMSM_ORD_REBUILD on
3475 * disks that are being rebuilt. New failures are recorded to
3476 * map[0]. So we look through all the disks we started with and
3477 * see if any failures are still present, or if any new ones
3478 * have arrived
3479 *
3480 * FIXME add support for online capacity expansion and
3481 * raid-level-migration
3482 */
3483 for (i = 0; i < prev->num_members; i++) {
3484 ord = __le32_to_cpu(prev->disk_ord_tbl[i]);
3485 ord |= __le32_to_cpu(map->disk_ord_tbl[i]);
3486 idx = ord_to_idx(ord);
3487
3488 disk = get_imsm_disk(super, idx);
3489 if (!disk || disk->status & FAILED_DISK ||
3490 ord & IMSM_ORD_REBUILD)
3491 failed++;
3492 }
3493
3494 return failed;
3495 }
3496
3497 static int is_resyncing(struct imsm_dev *dev)
3498 {
3499 struct imsm_map *migr_map;
3500
3501 if (!dev->vol.migr_state)
3502 return 0;
3503
3504 if (dev->vol.migr_type == MIGR_INIT)
3505 return 1;
3506
3507 migr_map = get_imsm_map(dev, 1);
3508
3509 if (migr_map->map_state == IMSM_T_STATE_NORMAL)
3510 return 1;
3511 else
3512 return 0;
3513 }
3514
3515 static int is_rebuilding(struct imsm_dev *dev)
3516 {
3517 struct imsm_map *migr_map;
3518
3519 if (!dev->vol.migr_state)
3520 return 0;
3521
3522 if (dev->vol.migr_type != MIGR_REBUILD)
3523 return 0;
3524
3525 migr_map = get_imsm_map(dev, 1);
3526
3527 if (migr_map->map_state == IMSM_T_STATE_DEGRADED)
3528 return 1;
3529 else
3530 return 0;
3531 }
3532
3533 /* return true if we recorded new information */
3534 static int mark_failure(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3535 {
3536 __u32 ord;
3537 int slot;
3538 struct imsm_map *map;
3539
3540 /* new failures are always set in map[0] */
3541 map = get_imsm_map(dev, 0);
3542
3543 slot = get_imsm_disk_slot(map, idx);
3544 if (slot < 0)
3545 return 0;
3546
3547 ord = __le32_to_cpu(map->disk_ord_tbl[slot]);
3548 if ((disk->status & FAILED_DISK) && (ord & IMSM_ORD_REBUILD))
3549 return 0;
3550
3551 disk->status |= FAILED_DISK;
3552 set_imsm_ord_tbl_ent(map, slot, idx | IMSM_ORD_REBUILD);
3553 if (map->failed_disk_num == ~0)
3554 map->failed_disk_num = slot;
3555 return 1;
3556 }
3557
3558 static void mark_missing(struct imsm_dev *dev, struct imsm_disk *disk, int idx)
3559 {
3560 mark_failure(dev, disk, idx);
3561
3562 if (disk->scsi_id == __cpu_to_le32(~(__u32)0))
3563 return;
3564
3565 disk->scsi_id = __cpu_to_le32(~(__u32)0);
3566 memmove(&disk->serial[0], &disk->serial[1], MAX_RAID_SERIAL_LEN - 1);
3567 }
3568
3569 /* Handle dirty -> clean transititions and resync. Degraded and rebuild
3570 * states are handled in imsm_set_disk() with one exception, when a
3571 * resync is stopped due to a new failure this routine will set the
3572 * 'degraded' state for the array.
3573 */
3574 static int imsm_set_array_state(struct active_array *a, int consistent)
3575 {
3576 int inst = a->info.container_member;
3577 struct intel_super *super = a->container->sb;
3578 struct imsm_dev *dev = get_imsm_dev(super, inst);
3579 struct imsm_map *map = get_imsm_map(dev, 0);
3580 int failed = imsm_count_failed(super, dev);
3581 __u8 map_state = imsm_check_degraded(super, dev, failed);
3582
3583 /* before we activate this array handle any missing disks */
3584 if (consistent == 2 && super->missing) {
3585 struct dl *dl;
3586
3587 dprintf("imsm: mark missing\n");
3588 end_migration(dev, map_state);
3589 for (dl = super->missing; dl; dl = dl->next)
3590 mark_missing(dev, &dl->disk, dl->index);
3591 super->updates_pending++;
3592 }
3593
3594 if (consistent == 2 &&
3595 (!is_resync_complete(a) ||
3596 map_state != IMSM_T_STATE_NORMAL ||
3597 dev->vol.migr_state))
3598 consistent = 0;
3599
3600 if (is_resync_complete(a)) {
3601 /* complete intialization / resync,
3602 * recovery and interrupted recovery is completed in
3603 * ->set_disk
3604 */
3605 if (is_resyncing(dev)) {
3606 dprintf("imsm: mark resync done\n");
3607 end_migration(dev, map_state);
3608 super->updates_pending++;
3609 }
3610 } else if (!is_resyncing(dev) && !failed) {
3611 /* mark the start of the init process if nothing is failed */
3612 dprintf("imsm: mark resync start (%llu)\n", a->resync_start);
3613 if (map->map_state == IMSM_T_STATE_NORMAL)
3614 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_REBUILD);
3615 else
3616 migrate(dev, IMSM_T_STATE_NORMAL, MIGR_INIT);
3617 super->updates_pending++;
3618 }
3619
3620 /* check if we can update the migration checkpoint */
3621 if (dev->vol.migr_state &&
3622 __le32_to_cpu(dev->vol.curr_migr_unit) != a->resync_start) {
3623 dprintf("imsm: checkpoint migration (%llu)\n", a->resync_start);
3624 dev->vol.curr_migr_unit = __cpu_to_le32(a->resync_start);
3625 super->updates_pending++;
3626 }
3627
3628 /* mark dirty / clean */
3629 if (dev->vol.dirty != !consistent) {
3630 dprintf("imsm: mark '%s' (%llu)\n",
3631 consistent ? "clean" : "dirty", a->resync_start);
3632 if (consistent)
3633 dev->vol.dirty = 0;
3634 else
3635 dev->vol.dirty = 1;
3636 super->updates_pending++;
3637 }
3638 return consistent;
3639 }
3640
3641 static void imsm_set_disk(struct active_array *a, int n, int state)
3642 {
3643 int inst = a->info.container_member;
3644 struct intel_super *super = a->container->sb;
3645 struct imsm_dev *dev = get_imsm_dev(super, inst);
3646 struct imsm_map *map = get_imsm_map(dev, 0);
3647 struct imsm_disk *disk;
3648 int failed;
3649 __u32 ord;
3650 __u8 map_state;
3651
3652 if (n > map->num_members)
3653 fprintf(stderr, "imsm: set_disk %d out of range 0..%d\n",
3654 n, map->num_members - 1);
3655
3656 if (n < 0)
3657 return;
3658
3659 dprintf("imsm: set_disk %d:%x\n", n, state);
3660
3661 ord = get_imsm_ord_tbl_ent(dev, n);
3662 disk = get_imsm_disk(super, ord_to_idx(ord));
3663
3664 /* check for new failures */
3665 if (state & DS_FAULTY) {
3666 if (mark_failure(dev, disk, ord_to_idx(ord)))
3667 super->updates_pending++;
3668 }
3669
3670 /* check if in_sync */
3671 if (state & DS_INSYNC && ord & IMSM_ORD_REBUILD && is_rebuilding(dev)) {
3672 struct imsm_map *migr_map = get_imsm_map(dev, 1);
3673
3674 set_imsm_ord_tbl_ent(migr_map, n, ord_to_idx(ord));
3675 super->updates_pending++;
3676 }
3677
3678 failed = imsm_count_failed(super, dev);
3679 map_state = imsm_check_degraded(super, dev, failed);
3680
3681 /* check if recovery complete, newly degraded, or failed */
3682 if (map_state == IMSM_T_STATE_NORMAL && is_rebuilding(dev)) {
3683 end_migration(dev, map_state);
3684 map = get_imsm_map(dev, 0);
3685 map->failed_disk_num = ~0;
3686 super->updates_pending++;
3687 } else if (map_state == IMSM_T_STATE_DEGRADED &&
3688 map->map_state != map_state &&
3689 !dev->vol.migr_state) {
3690 dprintf("imsm: mark degraded\n");
3691 map->map_state = map_state;
3692 super->updates_pending++;
3693 } else if (map_state == IMSM_T_STATE_FAILED &&
3694 map->map_state != map_state) {
3695 dprintf("imsm: mark failed\n");
3696 end_migration(dev, map_state);
3697 super->updates_pending++;
3698 }
3699 }
3700
3701 static int store_imsm_mpb(int fd, struct intel_super *super)
3702 {
3703 struct imsm_super *mpb = super->anchor;
3704 __u32 mpb_size = __le32_to_cpu(mpb->mpb_size);
3705 unsigned long long dsize;
3706 unsigned long long sectors;
3707
3708 get_dev_size(fd, NULL, &dsize);
3709
3710 if (mpb_size > 512) {
3711 /* -1 to account for anchor */
3712 sectors = mpb_sectors(mpb) - 1;
3713
3714 /* write the extended mpb to the sectors preceeding the anchor */
3715 if (lseek64(fd, dsize - (512 * (2 + sectors)), SEEK_SET) < 0)
3716 return 1;
3717
3718 if (write(fd, super->buf + 512, 512 * sectors) != 512 * sectors)
3719 return 1;
3720 }
3721
3722 /* first block is stored on second to last sector of the disk */
3723 if (lseek64(fd, dsize - (512 * 2), SEEK_SET) < 0)
3724 return 1;
3725
3726 if (write(fd, super->buf, 512) != 512)
3727 return 1;
3728
3729 return 0;
3730 }
3731
3732 static void imsm_sync_metadata(struct supertype *container)
3733 {
3734 struct intel_super *super = container->sb;
3735
3736 if (!super->updates_pending)
3737 return;
3738
3739 write_super_imsm(super, 0);
3740
3741 super->updates_pending = 0;
3742 }
3743
3744 static struct dl *imsm_readd(struct intel_super *super, int idx, struct active_array *a)
3745 {
3746 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3747 int i = get_imsm_disk_idx(dev, idx);
3748 struct dl *dl;
3749
3750 for (dl = super->disks; dl; dl = dl->next)
3751 if (dl->index == i)
3752 break;
3753
3754 if (dl && dl->disk.status & FAILED_DISK)
3755 dl = NULL;
3756
3757 if (dl)
3758 dprintf("%s: found %x:%x\n", __func__, dl->major, dl->minor);
3759
3760 return dl;
3761 }
3762
3763 static struct dl *imsm_add_spare(struct intel_super *super, int slot,
3764 struct active_array *a, int activate_new)
3765 {
3766 struct imsm_dev *dev = get_imsm_dev(super, a->info.container_member);
3767 int idx = get_imsm_disk_idx(dev, slot);
3768 struct imsm_super *mpb = super->anchor;
3769 struct imsm_map *map;
3770 unsigned long long esize;
3771 unsigned long long pos;
3772 struct mdinfo *d;
3773 struct extent *ex;
3774 int i, j;
3775 int found;
3776 __u32 array_start;
3777 __u32 blocks;
3778 struct dl *dl;
3779
3780 for (dl = super->disks; dl; dl = dl->next) {
3781 /* If in this array, skip */
3782 for (d = a->info.devs ; d ; d = d->next)
3783 if (d->state_fd >= 0 &&
3784 d->disk.major == dl->major &&
3785 d->disk.minor == dl->minor) {
3786 dprintf("%x:%x already in array\n", dl->major, dl->minor);
3787 break;
3788 }
3789 if (d)
3790 continue;
3791
3792 /* skip in use or failed drives */
3793 if (dl->disk.status & FAILED_DISK || idx == dl->index ||
3794 dl->index == -2) {
3795 dprintf("%x:%x status (failed: %d index: %d)\n",
3796 dl->major, dl->minor,
3797 (dl->disk.status & FAILED_DISK) == FAILED_DISK, idx);
3798 continue;
3799 }
3800
3801 /* skip pure spares when we are looking for partially
3802 * assimilated drives
3803 */
3804 if (dl->index == -1 && !activate_new)
3805 continue;
3806
3807 /* Does this unused device have the requisite free space?
3808 * It needs to be able to cover all member volumes
3809 */
3810 ex = get_extents(super, dl);
3811 if (!ex) {
3812 dprintf("cannot get extents\n");
3813 continue;
3814 }
3815 for (i = 0; i < mpb->num_raid_devs; i++) {
3816 dev = get_imsm_dev(super, i);
3817 map = get_imsm_map(dev, 0);
3818
3819 /* check if this disk is already a member of
3820 * this array
3821 */
3822 if (get_imsm_disk_slot(map, dl->index) >= 0)
3823 continue;
3824
3825 found = 0;
3826 j = 0;
3827 pos = 0;
3828 array_start = __le32_to_cpu(map->pba_of_lba0);
3829 blocks = __le32_to_cpu(map->blocks_per_member);
3830
3831 do {
3832 /* check that we can start at pba_of_lba0 with
3833 * blocks_per_member of space
3834 */
3835 esize = ex[j].start - pos;
3836 if (array_start >= pos &&
3837 array_start + blocks < ex[j].start) {
3838 found = 1;
3839 break;
3840 }
3841 pos = ex[j].start + ex[j].size;
3842 j++;
3843 } while (ex[j-1].size);
3844
3845 if (!found)
3846 break;
3847 }
3848
3849 free(ex);
3850 if (i < mpb->num_raid_devs) {
3851 dprintf("%x:%x does not have %u at %u\n",
3852 dl->major, dl->minor,
3853 blocks, array_start);
3854 /* No room */
3855 continue;
3856 }
3857 return dl;
3858 }
3859
3860 return dl;
3861 }
3862
3863 static struct mdinfo *imsm_activate_spare(struct active_array *a,
3864 struct metadata_update **updates)
3865 {
3866 /**
3867 * Find a device with unused free space and use it to replace a
3868 * failed/vacant region in an array. We replace failed regions one a
3869 * array at a time. The result is that a new spare disk will be added
3870 * to the first failed array and after the monitor has finished
3871 * propagating failures the remainder will be consumed.
3872 *
3873 * FIXME add a capability for mdmon to request spares from another
3874 * container.
3875 */
3876
3877 struct intel_super *super = a->container->sb;
3878 int inst = a->info.container_member;
3879 struct imsm_dev *dev = get_imsm_dev(super, inst);
3880 struct imsm_map *map = get_imsm_map(dev, 0);
3881 int failed = a->info.array.raid_disks;
3882 struct mdinfo *rv = NULL;
3883 struct mdinfo *d;
3884 struct mdinfo *di;
3885 struct metadata_update *mu;
3886 struct dl *dl;
3887 struct imsm_update_activate_spare *u;
3888 int num_spares = 0;
3889 int i;
3890
3891 for (d = a->info.devs ; d ; d = d->next) {
3892 if ((d->curr_state & DS_FAULTY) &&
3893 d->state_fd >= 0)
3894 /* wait for Removal to happen */
3895 return NULL;
3896 if (d->state_fd >= 0)
3897 failed--;
3898 }
3899
3900 dprintf("imsm: activate spare: inst=%d failed=%d (%d) level=%d\n",
3901 inst, failed, a->info.array.raid_disks, a->info.array.level);
3902 if (imsm_check_degraded(super, dev, failed) != IMSM_T_STATE_DEGRADED)
3903 return NULL;
3904
3905 /* For each slot, if it is not working, find a spare */
3906 for (i = 0; i < a->info.array.raid_disks; i++) {
3907 for (d = a->info.devs ; d ; d = d->next)
3908 if (d->disk.raid_disk == i)
3909 break;
3910 dprintf("found %d: %p %x\n", i, d, d?d->curr_state:0);
3911 if (d && (d->state_fd >= 0))
3912 continue;
3913
3914 /*
3915 * OK, this device needs recovery. Try to re-add the
3916 * previous occupant of this slot, if this fails see if
3917 * we can continue the assimilation of a spare that was
3918 * partially assimilated, finally try to activate a new
3919 * spare.
3920 */
3921 dl = imsm_readd(super, i, a);
3922 if (!dl)
3923 dl = imsm_add_spare(super, i, a, 0);
3924 if (!dl)
3925 dl = imsm_add_spare(super, i, a, 1);
3926 if (!dl)
3927 continue;
3928
3929 /* found a usable disk with enough space */
3930 di = malloc(sizeof(*di));
3931 if (!di)
3932 continue;
3933 memset(di, 0, sizeof(*di));
3934
3935 /* dl->index will be -1 in the case we are activating a
3936 * pristine spare. imsm_process_update() will create a
3937 * new index in this case. Once a disk is found to be
3938 * failed in all member arrays it is kicked from the
3939 * metadata
3940 */
3941 di->disk.number = dl->index;
3942
3943 /* (ab)use di->devs to store a pointer to the device
3944 * we chose
3945 */
3946 di->devs = (struct mdinfo *) dl;
3947
3948 di->disk.raid_disk = i;
3949 di->disk.major = dl->major;
3950 di->disk.minor = dl->minor;
3951 di->disk.state = 0;
3952 di->data_offset = __le32_to_cpu(map->pba_of_lba0);
3953 di->component_size = a->info.component_size;
3954 di->container_member = inst;
3955 di->next = rv;
3956 rv = di;
3957 num_spares++;
3958 dprintf("%x:%x to be %d at %llu\n", dl->major, dl->minor,
3959 i, di->data_offset);
3960
3961 break;
3962 }
3963
3964 if (!rv)
3965 /* No spares found */
3966 return rv;
3967 /* Now 'rv' has a list of devices to return.
3968 * Create a metadata_update record to update the
3969 * disk_ord_tbl for the array
3970 */
3971 mu = malloc(sizeof(*mu));
3972 if (mu) {
3973 mu->buf = malloc(sizeof(struct imsm_update_activate_spare) * num_spares);
3974 if (mu->buf == NULL) {
3975 free(mu);
3976 mu = NULL;
3977 }
3978 }
3979 if (!mu) {
3980 while (rv) {
3981 struct mdinfo *n = rv->next;
3982
3983 free(rv);
3984 rv = n;
3985 }
3986 return NULL;
3987 }
3988
3989 mu->space = NULL;
3990 mu->len = sizeof(struct imsm_update_activate_spare) * num_spares;
3991 mu->next = *updates;
3992 u = (struct imsm_update_activate_spare *) mu->buf;
3993
3994 for (di = rv ; di ; di = di->next) {
3995 u->type = update_activate_spare;
3996 u->dl = (struct dl *) di->devs;
3997 di->devs = NULL;
3998 u->slot = di->disk.raid_disk;
3999 u->array = inst;
4000 u->next = u + 1;
4001 u++;
4002 }
4003 (u-1)->next = NULL;
4004 *updates = mu;
4005
4006 return rv;
4007 }
4008
4009 static int disks_overlap(struct intel_super *super, int idx, struct imsm_update_create_array *u)
4010 {
4011 struct imsm_dev *dev = get_imsm_dev(super, idx);
4012 struct imsm_map *map = get_imsm_map(dev, 0);
4013 struct imsm_map *new_map = get_imsm_map(&u->dev, 0);
4014 struct disk_info *inf = get_disk_info(u);
4015 struct imsm_disk *disk;
4016 int i;
4017 int j;
4018
4019 for (i = 0; i < map->num_members; i++) {
4020 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4021 for (j = 0; j < new_map->num_members; j++)
4022 if (serialcmp(disk->serial, inf[j].serial) == 0)
4023 return 1;
4024 }
4025
4026 return 0;
4027 }
4028
4029 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index);
4030
4031 static void imsm_process_update(struct supertype *st,
4032 struct metadata_update *update)
4033 {
4034 /**
4035 * crack open the metadata_update envelope to find the update record
4036 * update can be one of:
4037 * update_activate_spare - a spare device has replaced a failed
4038 * device in an array, update the disk_ord_tbl. If this disk is
4039 * present in all member arrays then also clear the SPARE_DISK
4040 * flag
4041 */
4042 struct intel_super *super = st->sb;
4043 struct imsm_super *mpb;
4044 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4045
4046 /* update requires a larger buf but the allocation failed */
4047 if (super->next_len && !super->next_buf) {
4048 super->next_len = 0;
4049 return;
4050 }
4051
4052 if (super->next_buf) {
4053 memcpy(super->next_buf, super->buf, super->len);
4054 free(super->buf);
4055 super->len = super->next_len;
4056 super->buf = super->next_buf;
4057
4058 super->next_len = 0;
4059 super->next_buf = NULL;
4060 }
4061
4062 mpb = super->anchor;
4063
4064 switch (type) {
4065 case update_activate_spare: {
4066 struct imsm_update_activate_spare *u = (void *) update->buf;
4067 struct imsm_dev *dev = get_imsm_dev(super, u->array);
4068 struct imsm_map *map = get_imsm_map(dev, 0);
4069 struct imsm_map *migr_map;
4070 struct active_array *a;
4071 struct imsm_disk *disk;
4072 __u8 to_state;
4073 struct dl *dl;
4074 unsigned int found;
4075 int failed;
4076 int victim = get_imsm_disk_idx(dev, u->slot);
4077 int i;
4078
4079 for (dl = super->disks; dl; dl = dl->next)
4080 if (dl == u->dl)
4081 break;
4082
4083 if (!dl) {
4084 fprintf(stderr, "error: imsm_activate_spare passed "
4085 "an unknown disk (index: %d)\n",
4086 u->dl->index);
4087 return;
4088 }
4089
4090 super->updates_pending++;
4091
4092 /* count failures (excluding rebuilds and the victim)
4093 * to determine map[0] state
4094 */
4095 failed = 0;
4096 for (i = 0; i < map->num_members; i++) {
4097 if (i == u->slot)
4098 continue;
4099 disk = get_imsm_disk(super, get_imsm_disk_idx(dev, i));
4100 if (!disk || disk->status & FAILED_DISK)
4101 failed++;
4102 }
4103
4104 /* adding a pristine spare, assign a new index */
4105 if (dl->index < 0) {
4106 dl->index = super->anchor->num_disks;
4107 super->anchor->num_disks++;
4108 }
4109 disk = &dl->disk;
4110 disk->status |= CONFIGURED_DISK;
4111 disk->status &= ~SPARE_DISK;
4112
4113 /* mark rebuild */
4114 to_state = imsm_check_degraded(super, dev, failed);
4115 map->map_state = IMSM_T_STATE_DEGRADED;
4116 migrate(dev, to_state, MIGR_REBUILD);
4117 migr_map = get_imsm_map(dev, 1);
4118 set_imsm_ord_tbl_ent(map, u->slot, dl->index);
4119 set_imsm_ord_tbl_ent(migr_map, u->slot, dl->index | IMSM_ORD_REBUILD);
4120
4121 /* count arrays using the victim in the metadata */
4122 found = 0;
4123 for (a = st->arrays; a ; a = a->next) {
4124 dev = get_imsm_dev(super, a->info.container_member);
4125 map = get_imsm_map(dev, 0);
4126
4127 if (get_imsm_disk_slot(map, victim) >= 0)
4128 found++;
4129 }
4130
4131 /* delete the victim if it is no longer being
4132 * utilized anywhere
4133 */
4134 if (!found) {
4135 struct dl **dlp;
4136
4137 /* We know that 'manager' isn't touching anything,
4138 * so it is safe to delete
4139 */
4140 for (dlp = &super->disks; *dlp; dlp = &(*dlp)->next)
4141 if ((*dlp)->index == victim)
4142 break;
4143
4144 /* victim may be on the missing list */
4145 if (!*dlp)
4146 for (dlp = &super->missing; *dlp; dlp = &(*dlp)->next)
4147 if ((*dlp)->index == victim)
4148 break;
4149 imsm_delete(super, dlp, victim);
4150 }
4151 break;
4152 }
4153 case update_create_array: {
4154 /* someone wants to create a new array, we need to be aware of
4155 * a few races/collisions:
4156 * 1/ 'Create' called by two separate instances of mdadm
4157 * 2/ 'Create' versus 'activate_spare': mdadm has chosen
4158 * devices that have since been assimilated via
4159 * activate_spare.
4160 * In the event this update can not be carried out mdadm will
4161 * (FIX ME) notice that its update did not take hold.
4162 */
4163 struct imsm_update_create_array *u = (void *) update->buf;
4164 struct intel_dev *dv;
4165 struct imsm_dev *dev;
4166 struct imsm_map *map, *new_map;
4167 unsigned long long start, end;
4168 unsigned long long new_start, new_end;
4169 int i;
4170 struct disk_info *inf;
4171 struct dl *dl;
4172
4173 /* handle racing creates: first come first serve */
4174 if (u->dev_idx < mpb->num_raid_devs) {
4175 dprintf("%s: subarray %d already defined\n",
4176 __func__, u->dev_idx);
4177 goto create_error;
4178 }
4179
4180 /* check update is next in sequence */
4181 if (u->dev_idx != mpb->num_raid_devs) {
4182 dprintf("%s: can not create array %d expected index %d\n",
4183 __func__, u->dev_idx, mpb->num_raid_devs);
4184 goto create_error;
4185 }
4186
4187 new_map = get_imsm_map(&u->dev, 0);
4188 new_start = __le32_to_cpu(new_map->pba_of_lba0);
4189 new_end = new_start + __le32_to_cpu(new_map->blocks_per_member);
4190 inf = get_disk_info(u);
4191
4192 /* handle activate_spare versus create race:
4193 * check to make sure that overlapping arrays do not include
4194 * overalpping disks
4195 */
4196 for (i = 0; i < mpb->num_raid_devs; i++) {
4197 dev = get_imsm_dev(super, i);
4198 map = get_imsm_map(dev, 0);
4199 start = __le32_to_cpu(map->pba_of_lba0);
4200 end = start + __le32_to_cpu(map->blocks_per_member);
4201 if ((new_start >= start && new_start <= end) ||
4202 (start >= new_start && start <= new_end))
4203 /* overlap */;
4204 else
4205 continue;
4206
4207 if (disks_overlap(super, i, u)) {
4208 dprintf("%s: arrays overlap\n", __func__);
4209 goto create_error;
4210 }
4211 }
4212
4213 /* check that prepare update was successful */
4214 if (!update->space) {
4215 dprintf("%s: prepare update failed\n", __func__);
4216 goto create_error;
4217 }
4218
4219 /* check that all disks are still active before committing
4220 * changes. FIXME: could we instead handle this by creating a
4221 * degraded array? That's probably not what the user expects,
4222 * so better to drop this update on the floor.
4223 */
4224 for (i = 0; i < new_map->num_members; i++) {
4225 dl = serial_to_dl(inf[i].serial, super);
4226 if (!dl) {
4227 dprintf("%s: disk disappeared\n", __func__);
4228 goto create_error;
4229 }
4230 }
4231
4232 super->updates_pending++;
4233
4234 /* convert spares to members and fixup ord_tbl */
4235 for (i = 0; i < new_map->num_members; i++) {
4236 dl = serial_to_dl(inf[i].serial, super);
4237 if (dl->index == -1) {
4238 dl->index = mpb->num_disks;
4239 mpb->num_disks++;
4240 dl->disk.status |= CONFIGURED_DISK;
4241 dl->disk.status &= ~SPARE_DISK;
4242 }
4243 set_imsm_ord_tbl_ent(new_map, i, dl->index);
4244 }
4245
4246 dv = update->space;
4247 dev = dv->dev;
4248 update->space = NULL;
4249 imsm_copy_dev(dev, &u->dev);
4250 dv->index = u->dev_idx;
4251 dv->next = super->devlist;
4252 super->devlist = dv;
4253 mpb->num_raid_devs++;
4254
4255 imsm_update_version_info(super);
4256 break;
4257 create_error:
4258 /* mdmon knows how to release update->space, but not
4259 * ((struct intel_dev *) update->space)->dev
4260 */
4261 if (update->space) {
4262 dv = update->space;
4263 free(dv->dev);
4264 }
4265 break;
4266 }
4267 case update_add_disk:
4268
4269 /* we may be able to repair some arrays if disks are
4270 * being added */
4271 if (super->add) {
4272 struct active_array *a;
4273
4274 super->updates_pending++;
4275 for (a = st->arrays; a; a = a->next)
4276 a->check_degraded = 1;
4277 }
4278 /* add some spares to the metadata */
4279 while (super->add) {
4280 struct dl *al;
4281
4282 al = super->add;
4283 super->add = al->next;
4284 al->next = super->disks;
4285 super->disks = al;
4286 dprintf("%s: added %x:%x\n",
4287 __func__, al->major, al->minor);
4288 }
4289
4290 break;
4291 }
4292 }
4293
4294 static void imsm_prepare_update(struct supertype *st,
4295 struct metadata_update *update)
4296 {
4297 /**
4298 * Allocate space to hold new disk entries, raid-device entries or a new
4299 * mpb if necessary. The manager synchronously waits for updates to
4300 * complete in the monitor, so new mpb buffers allocated here can be
4301 * integrated by the monitor thread without worrying about live pointers
4302 * in the manager thread.
4303 */
4304 enum imsm_update_type type = *(enum imsm_update_type *) update->buf;
4305 struct intel_super *super = st->sb;
4306 struct imsm_super *mpb = super->anchor;
4307 size_t buf_len;
4308 size_t len = 0;
4309
4310 switch (type) {
4311 case update_create_array: {
4312 struct imsm_update_create_array *u = (void *) update->buf;
4313 struct intel_dev *dv;
4314 struct imsm_dev *dev = &u->dev;
4315 struct imsm_map *map = get_imsm_map(dev, 0);
4316 struct dl *dl;
4317 struct disk_info *inf;
4318 int i;
4319 int activate = 0;
4320
4321 inf = get_disk_info(u);
4322 len = sizeof_imsm_dev(dev, 1);
4323 /* allocate a new super->devlist entry */
4324 dv = malloc(sizeof(*dv));
4325 if (dv) {
4326 dv->dev = malloc(len);
4327 if (dv->dev)
4328 update->space = dv;
4329 else {
4330 free(dv);
4331 update->space = NULL;
4332 }
4333 }
4334
4335 /* count how many spares will be converted to members */
4336 for (i = 0; i < map->num_members; i++) {
4337 dl = serial_to_dl(inf[i].serial, super);
4338 if (!dl) {
4339 /* hmm maybe it failed?, nothing we can do about
4340 * it here
4341 */
4342 continue;
4343 }
4344 if (count_memberships(dl, super) == 0)
4345 activate++;
4346 }
4347 len += activate * sizeof(struct imsm_disk);
4348 break;
4349 default:
4350 break;
4351 }
4352 }
4353
4354 /* check if we need a larger metadata buffer */
4355 if (super->next_buf)
4356 buf_len = super->next_len;
4357 else
4358 buf_len = super->len;
4359
4360 if (__le32_to_cpu(mpb->mpb_size) + len > buf_len) {
4361 /* ok we need a larger buf than what is currently allocated
4362 * if this allocation fails process_update will notice that
4363 * ->next_len is set and ->next_buf is NULL
4364 */
4365 buf_len = ROUND_UP(__le32_to_cpu(mpb->mpb_size) + len, 512);
4366 if (super->next_buf)
4367 free(super->next_buf);
4368
4369 super->next_len = buf_len;
4370 if (posix_memalign(&super->next_buf, 512, buf_len) != 0)
4371 super->next_buf = NULL;
4372 }
4373 }
4374
4375 /* must be called while manager is quiesced */
4376 static void imsm_delete(struct intel_super *super, struct dl **dlp, int index)
4377 {
4378 struct imsm_super *mpb = super->anchor;
4379 struct dl *iter;
4380 struct imsm_dev *dev;
4381 struct imsm_map *map;
4382 int i, j, num_members;
4383 __u32 ord;
4384
4385 dprintf("%s: deleting device[%d] from imsm_super\n",
4386 __func__, index);
4387
4388 /* shift all indexes down one */
4389 for (iter = super->disks; iter; iter = iter->next)
4390 if (iter->index > index)
4391 iter->index--;
4392 for (iter = super->missing; iter; iter = iter->next)
4393 if (iter->index > index)
4394 iter->index--;
4395
4396 for (i = 0; i < mpb->num_raid_devs; i++) {
4397 dev = get_imsm_dev(super, i);
4398 map = get_imsm_map(dev, 0);
4399 num_members = map->num_members;
4400 for (j = 0; j < num_members; j++) {
4401 /* update ord entries being careful not to propagate
4402 * ord-flags to the first map
4403 */
4404 ord = get_imsm_ord_tbl_ent(dev, j);
4405
4406 if (ord_to_idx(ord) <= index)
4407 continue;
4408
4409 map = get_imsm_map(dev, 0);
4410 set_imsm_ord_tbl_ent(map, j, ord_to_idx(ord - 1));
4411 map = get_imsm_map(dev, 1);
4412 if (map)
4413 set_imsm_ord_tbl_ent(map, j, ord - 1);
4414 }
4415 }
4416
4417 mpb->num_disks--;
4418 super->updates_pending++;
4419 if (*dlp) {
4420 struct dl *dl = *dlp;
4421
4422 *dlp = (*dlp)->next;
4423 __free_imsm_disk(dl);
4424 }
4425 }
4426 #endif /* MDASSEMBLE */
4427
4428 struct superswitch super_imsm = {
4429 #ifndef MDASSEMBLE
4430 .examine_super = examine_super_imsm,
4431 .brief_examine_super = brief_examine_super_imsm,
4432 .detail_super = detail_super_imsm,
4433 .brief_detail_super = brief_detail_super_imsm,
4434 .write_init_super = write_init_super_imsm,
4435 .validate_geometry = validate_geometry_imsm,
4436 .add_to_super = add_to_super_imsm,
4437 .detail_platform = detail_platform_imsm,
4438 #endif
4439 .match_home = match_home_imsm,
4440 .uuid_from_super= uuid_from_super_imsm,
4441 .getinfo_super = getinfo_super_imsm,
4442 .update_super = update_super_imsm,
4443
4444 .avail_size = avail_size_imsm,
4445
4446 .compare_super = compare_super_imsm,
4447
4448 .load_super = load_super_imsm,
4449 .init_super = init_super_imsm,
4450 .store_super = store_zero_imsm,
4451 .free_super = free_super_imsm,
4452 .match_metadata_desc = match_metadata_desc_imsm,
4453 .container_content = container_content_imsm,
4454 .default_layout = imsm_level_to_layout,
4455
4456 .external = 1,
4457 .name = "imsm",
4458
4459 #ifndef MDASSEMBLE
4460 /* for mdmon */
4461 .open_new = imsm_open_new,
4462 .load_super = load_super_imsm,
4463 .set_array_state= imsm_set_array_state,
4464 .set_disk = imsm_set_disk,
4465 .sync_metadata = imsm_sync_metadata,
4466 .activate_spare = imsm_activate_spare,
4467 .process_update = imsm_process_update,
4468 .prepare_update = imsm_prepare_update,
4469 #endif /* MDASSEMBLE */
4470 };
Unnamed repository; edit this file 'description' to name the repository.